Special unmanned aerial vehicle cruising relay data transmission system for power inspection

Abstract

The application discloses a special unmanned aerial vehicle cruising relay data transmission system for power inspection, relates to the technical field of data transmission, and comprises the following steps: acquiring the inspection position and inspection data transmitted by the unmanned aerial vehicle in power inspection, and calling the position power characteristics; performing inspection data quality influence prediction, obtaining a predicted data quality influence coefficient, an actual data quality influence coefficient, and a data quality coefficient; analyzing and obtaining an inspection position deviation coefficient; performing inspection data generation network selection optimization, obtaining an optimized generation network group, generating inspection data, and performing relay transmission. The application solves the technical problem that the inspection data of the unmanned aerial vehicle in the existing power inspection is affected by electromagnetic interference and the data quality is low after relay transmission.

Description

Technical Field

[0001] This application relates to the field of data transmission technology, specifically to a relay data transmission system for power line inspection drone patrols. Background Technology

[0002] With the rapid development of the power industry and the continuous expansion of the power grid, inspection tasks are becoming increasingly heavy. Drones, equipped with high-definition cameras, infrared thermal imagers, and other sensors, can take close-up, multi-angle photos and collect data on transmission lines and substation equipment, promptly detecting potential faults such as line icing, tower tilting, and equipment overheating. They have been widely used in the field of power inspection.

[0003] However, during power line inspection by drones, the wireless communication link between the drone and the ground control center is easily affected by factors such as obstruction and electromagnetic interference, which can lead to problems such as packet loss, delay, data distortion, or even transmission interruption during data transmission.

[0004] Furthermore, the relay transmission system for UAV inspection data often fails to fully consider the impact of the power characteristics of the inspection location on data quality, making it difficult to dynamically select and optimize the network based on actual data quality and location deviations. This affects the reliability of the inspection data and the accuracy of subsequent analysis and decision-making. Summary of the Invention

[0005] This application provides a dedicated UAV patrol relay data transmission system for power line inspection, which solves the technical problem that the inspection data from UAVs in existing power line inspections is affected by electromagnetic interference, resulting in low data quality after relay transmission.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows: This application provides a dedicated UAV patrol relay data transmission system for power line inspection, including: The inspection information acquisition module is used to acquire the inspection location and inspection data transmitted by the drone during power inspection, and retrieve the location power characteristics based on the inspection location. The quality impact prediction module is used to predict the impact of inspection data quality based on the location power characteristics, obtain the predicted data quality impact coefficient, obtain the actual data quality impact coefficient based on the inspection data processing, and calculate the data quality coefficient. The deviation coefficient acquisition module is used to analyze and obtain the inspection position deviation coefficient based on the data quality coefficient. The inspection data optimization module is used to optimize the selection of the inspection data generation network according to the data quality coefficient and the inspection position deviation coefficient, obtain an optimized generation network group, input the inspection data, obtain generated inspection data, and perform relay transmission. The optimization adjustment direction is set according to the data quality coefficient.

[0007] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a dedicated UAV patrol relay data transmission system for power line inspection. First, it acquires the inspection location and data transmitted by the UAV and retrieves the corresponding power characteristics, providing foundational data for subsequent data quality assessment and network optimization. Second, it predicts the impact of location power characteristics on data quality and, combined with actual inspection data processing, obtains the actual impact coefficient. The difference between the two is calculated to obtain the data quality coefficient, providing a quantitative indicator for judging data reliability. Third, it analyzes the deviation of the inspection location based on the data quality coefficient. By randomly selecting other inspection locations and calculating their matching degree with the current location, it ultimately determines the inspection location deviation coefficient, identifying potential data acquisition problems due to location factors. Finally, it selects and optimizes the inspection data generation network. Through a preset number of random selections, similarity analysis, fitness assessment, and optimization direction and adjustment ratio settings based on the sign and absolute value of the data quality coefficient, iterative optimization is performed to obtain the optimal generation network group. The original inspection data is input into this network group to generate high-quality inspection data for relay transmission. This dynamic network selection and optimization adjustment mechanism can adapt to different transmission network environments based on real-time data quality and location deviation.

[0008] Through the above technical solution, this application improves the reliability of inspection data during transmission, ensures the accuracy of analysis and decision-making based on inspection data, and thus guarantees the efficient implementation of power inspection work. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the structure of the power inspection drone patrol relay data transmission system provided in the embodiments of this application.

[0011] The components represented by each number in the attached diagram are explained below: Inspection information collection module 11, quality impact prediction module 12, deviation coefficient acquisition module 13, and inspection data optimization module 14. Detailed Implementation

[0012] This application provides a dedicated drone patrol relay data transmission system for power line inspection, which addresses the technical problem that the inspection data from drones in existing power line inspections is affected by electromagnetic interference, resulting in low data quality after relay transmission.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0016] Examples, such as Figure 1 As shown in the embodiment of this application, a dedicated UAV patrol relay data transmission system for power line inspection is provided, including: The inspection information acquisition module 11 is used to acquire the inspection location and inspection data transmitted by the drone during power inspection, and retrieve the location power characteristics based on the inspection location. In this embodiment of the application, the latitude and longitude coordinates of the inspection location are obtained in real time through the positioning system carried by the UAV, and the inspection data collected and transmitted back by the UAV through the sensors during the inspection process are received. The inspection data includes inspection images, videos, temperature, equipment status parameter data, etc.

[0017] Based on the obtained latitude and longitude coordinates of the inspection location, the system accesses a pre-set power geographic information database or power feature map to retrieve the corresponding location's power characteristics. These location power characteristics specifically include the voltage level of the transmission lines in the area, the presence of strong electromagnetic interference sources, and historical electromagnetic environment monitoring data.

[0018] Specifically, the inspection information collection module 11 in the system includes: During the process of drone power line inspection, the inspection data transmitted by the drone and the drone's current inspection location are acquired. Based on the inspection location, the corresponding location power characteristics within the power network are obtained through indexing.

[0019] In this embodiment of the application, firstly, when the drone is performing a power inspection task, it receives various types of inspection data back in real time, including images of power transmission lines and equipment captured by a high-definition camera, temperature distribution data collected by an infrared thermal imager, and equipment operating status parameters monitored by various sensors.

[0020] Simultaneously, the drone's current inspection location coordinates are obtained via its GPS positioning system. Based on these coordinates, the system indexes and matches them in a pre-built power network geographic information system (GIS) database to retrieve the corresponding power characteristics of that coordinate point within the power network.

[0021] Furthermore, the location-specific power characteristics specifically include the rated voltage level of the main transmission lines in the area, such as 110kV; the density of the lines; the presence of large substations, converter stations, or other strong electromagnetic interference sources in the vicinity; and historical electromagnetic radiation intensity monitoring data and radio wave propagation loss model parameters for the same period in the area. The location-specific power characteristics reflect the power environment conditions at the inspection location. The higher the transmission line voltage or electromagnetic intensity, the greater the impact on the inspection data.

[0022] Specifically, based on the inspection location, the corresponding location power characteristics within the power network are indexed and obtained, including: Based on the sensors in the power network, the distribution of monitored power characteristics within the inspection area is obtained; Based on the inspection location, the location power features are extracted within the power feature distribution.

[0023] In this embodiment, firstly, various sensors deployed in the power network, such as current sensors, voltage sensors, and electromagnetic environment monitoring sensors on transmission lines, collect power parameter data within the inspection area, including line current, voltage values, electromagnetic radiation intensity, electric field strength, and magnetic field strength in the surrounding space. The collected data is then integrated according to geographical location information to construct a power characteristic distribution map of the inspection area. This power characteristic distribution map, based on geographical coordinates, reflects the magnitude and variation of power characteristic parameters at different locations.

[0024] Then, the obtained drone inspection location coordinates are used to locate the position within the constructed power characteristic distribution map. A spatial interpolation algorithm is then used to extract the specific power characteristic parameters corresponding to the inspection location from the map, i.e., the location power characteristics. The spatial interpolation algorithm can estimate unknown location data based on known data points; for example, it can extract the location power characteristics corresponding to the inspection location from the power characteristic distribution map.

[0025] The quality impact prediction module 12 is used to predict the impact of inspection data quality based on the location power characteristics, obtain the predicted data quality impact coefficient, obtain the actual data quality impact coefficient based on the inspection data processing, and calculate the data quality coefficient. In this embodiment, firstly, the data quality impact coefficient is predicted based on the power characteristics of the inspection location, such as the proportion of noise in the inspection image. Then, based on the inspection data, the actual data quality impact coefficient is obtained, and the difference between the two is calculated as the data quality coefficient.

[0026] Specifically, the quality impact prediction module 12 in the system includes: The location power characteristics are input into the data quality impact predictor, and the predicted data quality impact coefficient is output. Based on the inspection data, the actual data quality impact coefficient is obtained, wherein the actual data quality impact coefficient includes the proportion of noisy data points in the inspection data; The difference between the actual data quality impact coefficient and the predicted data quality impact coefficient is calculated to obtain the data quality coefficient.

[0027] In this embodiment, location power characteristics, such as transmission line voltage levels, intensity of strong electromagnetic interference sources, and historical electromagnetic environment monitoring data, are first input into a pre-trained data quality impact predictor. This predictor is built based on a deep learning model, trained using sample data of historical location power characteristics and corresponding data quality impact levels. It can output a quantified predicted data quality impact coefficient based on the input location power characteristics. This coefficient reflects the expected degree of interference to the inspection data quality under the influence of the current location power characteristics; for example, a larger value indicates more severe predicted interference and potentially lower data quality.

[0028] Secondly, the acquired actual inspection data is processed to obtain the actual data quality impact coefficient. Specifically, for inspection image data, image quality assessment algorithms are used, such as calculating the signal-to-noise ratio and mean square error, to statistically analyze the proportion of noise data points in the image, such as salt-and-pepper noise and Gaussian noise caused by electromagnetic interference in the entire image pixels. For temperature data or equipment status parameter data, the proportion of noise data points, such as jump values ​​exceeding the normal range and irregular abrupt changes, is determined by analyzing the fluctuation range of the data and the frequency of outliers.

[0029] Furthermore, the proportion of noise data points obtained from the statistics is used as the actual data quality impact coefficient, which reflects the quality status of the current inspection data after being affected by actual interference.

[0030] Finally, the difference between the actual data quality impact coefficient and the predicted data quality impact coefficient is calculated, i.e., data quality coefficient = actual data quality impact coefficient - predicted data quality impact coefficient. This data quality coefficient quantifies the deviation between the actual data quality and the predicted data quality. If the data quality coefficient is positive, it indicates that the actual data is more severely affected by interference than predicted; if it is negative, it indicates that the actual data quality is better than the predicted level; if it is close to zero, it indicates that the prediction is relatively accurate.

[0031] The training steps for the data quality impact predictor include: Based on the test records of the electromagnetic environment affecting the transmission data in the power area, a set of sample power characteristics was collected, and the data quality impact coefficients under different sample power characteristics were collected, labeled, and a set of sample predicted data quality impact coefficients was obtained. Based on machine learning, construct the structure of a data quality impact predictor; The data quality impact predictor is trained and tested under supervised supervision using the sample power feature set and the sample predicted data quality impact coefficient set as input data and labels until the test converges, thus completing the training.

[0032] In this embodiment, firstly, within the power area, tests are conducted to simulate scenarios involving different voltage levels, electromagnetic interference source intensities, and equipment operating states to assess the impact of the electromagnetic environment on transmitted data. Under each test scenario, power characteristic parameters are recorded, such as the voltage and current of the transmission line, the surrounding electromagnetic radiation intensity, and the type and distance of the interference source, forming a sample power characteristic set. Simultaneously, for the inspection data transmitted in each test, the aforementioned method for calculating the actual data quality impact coefficient is used to statistically analyze the proportion of noisy data points, which is then used as the label corresponding to that sample power characteristic, i.e., the sample predicted data quality impact coefficient set.

[0033] Secondly, based on machine learning algorithms, a Long Short-Term Memory (LSTM) network was chosen to construct the basic structure of the data quality impact predictor. The number of neurons in the input layer of the model corresponds to the dimension of the sample power features. For example, if there are three features: transmission line voltage level, intensity of strong electromagnetic interference sources, and historical electromagnetic environment monitoring data, then the input layer contains three nodes. The intermediate layers consist of 1-3 hidden layers, with the number of nodes in each layer adjusted experimentally, such as 64 or 32. The activation function used is ReLU. The output layer consists of one neuron, used to output the predicted data quality impact coefficient.

[0034] Then, the collected set of sample power features is used as the input data for the model, and the set of sample predicted data quality impact coefficients is used as the corresponding labels to supervise the training of the constructed data quality impact predictor. During training, the Adam optimizer and backpropagation algorithm are used to continuously adjust the model's weights and biases to minimize the loss function (e.g., mean squared error loss function) between the predicted output and the label. Simultaneously, the sample dataset is divided into training and test sets in a 7:3 ratio, and the model performance is periodically evaluated using the test set during training. When the loss function value on the test set stabilizes and reaches the preset convergence condition (e.g., the decrease in loss value is less than a set threshold over several consecutive training cycles), training stops. At this point, the data quality impact predictor is complete and has the ability to predict data quality impact coefficients based on the input location power features.

[0035] The deviation coefficient acquisition module 13 is used to analyze and obtain the inspection position deviation coefficient based on the data quality coefficient. In this embodiment of the application, due to electromagnetic interference, there may also be errors in the inspection position, so the inspection position deviation coefficient is obtained through analysis.

[0036] Specifically, the similarity between the predicted data quality coefficients of power characteristics at other locations and the current data quality coefficients is analyzed, the matching inspection location with the highest matching degree is selected, and then the inspection location deviation coefficient is calculated.

[0037] Specifically, the deviation coefficient acquisition module 13 in the system includes: Randomly select other inspection locations within the inspection area and retrieve the power characteristics of the random locations; The first random location matching degree is obtained by analyzing the power characteristics of random locations and the data quality coefficient. Continue to randomly select other inspection locations to obtain the matching inspection location with the highest location matching degree, calculate the distance to the inspection location, and combine it with the distance scale of the inspection area to calculate the inspection location deviation coefficient.

[0038] In this embodiment, firstly, within the inspection area, according to random sampling rules, multiple other potential inspection locations are randomly selected within the vicinity of the current inspection location of the UAV. For each randomly selected other inspection location, the corresponding random location power characteristics, such as the transmission line voltage level, electromagnetic interference source conditions, and historical electromagnetic environment data at that random location, are retrieved using the method described in the inspection information collection module.

[0039] Next, the random location power characteristics of each random location are input into the trained data quality impact predictor to obtain the predicted data quality impact coefficient corresponding to that random location. This predicted data quality impact coefficient is then compared with the actual data quality impact coefficient of the current inspection location, and the similarity between the two is calculated as the first random location matching degree between the random location and the current inspection location.

[0040] Specifically, the predicted data quality impact coefficient and the actual data quality impact coefficient at the current inspection location are converted into vector form, such as (0, a) and (0, b). Then, the cosine similarity method is used to calculate the similarity between the two, which is taken as the first random location matching degree. Furthermore, the matching degree is calculated using the cosine similarity method. The higher the similarity, the closer the prediction deviation caused by the power characteristics of the random location is to the current data quality coefficient, and the greater the probability that the random location is the actual inspection location.

[0041] Subsequently, other inspection locations are randomly selected using the same method described above, and their random location power characteristics, theoretical data quality coefficients, and matching degrees with the current data quality coefficients are repeatedly calculated. After obtaining a sufficient number of random location matching degrees, the random location with the highest matching degree is selected as the matching inspection location. This matching inspection location is the point that is closest to the actual inspection location of the UAV.

[0042] Finally, the straight-line distance between the current inspection location and the matched inspection location is calculated. Simultaneously, considering distance scales within the inspection area, such as the average density of power facilities and the typical error range of GPS positioning in that area, the above distance is normalized to obtain an inspection location deviation coefficient. A larger inspection location deviation coefficient indicates a potentially more severe positioning deviation.

[0043] For example, if the straight-line distance between the current inspection position and the matching inspection position is 50 meters and the maximum distance scale of the inspection area is 100 meters, then the inspection position deviation coefficient is the ratio of the two, which is 0.5.

[0044] Among them, the first random location matching degree is obtained by analyzing the power characteristics of random locations and data quality coefficients, including: The random location power characteristics are input into the data quality impact predictor, which outputs the random predicted data quality impact coefficient. Combined with the actual data quality impact coefficient, the first random data quality coefficient is calculated. Calculate the similarity between the first random data quality coefficient and the data quality coefficient, and use it as the first random location matching degree.

[0045] In this embodiment of the application, firstly, the power characteristics of the random location are input into the pre-trained data quality impact predictor, and the predictor outputs the random predicted data quality impact coefficient corresponding to the random location.

[0046] Secondly, based on the actual data quality impact coefficient obtained from the current inspection data processing, and according to the calculation formula of the data quality coefficient, namely, data quality coefficient = actual data quality impact coefficient - predicted data quality impact coefficient, the first random data quality coefficient corresponding to the random location is calculated.

[0047] Subsequently, cosine similarity is used to calculate the similarity between the first random data quality coefficient and the original data quality coefficient of the current inspection location. This similarity value serves as the first random location matching degree, measuring the degree of matching between the random location and the power characteristics of the current actual inspection location. The closer the similarity value is to 1, the higher the matching degree, and the more likely the random location is to be the accurate actual location of the UAV.

[0048] Furthermore, when calculating similarity, the current data quality coefficient and the first random data quality coefficient can be regarded as two vectors, and the similarity can be obtained by calculating the cosine angle between the vectors.

[0049] For example, if the current data quality coefficient is 0.3 and the first random data quality coefficient at a certain random position is 0.28, the two values ​​are close and the cosine similarity is close to 1, indicating that the matching degree of the random position is high; if the first random data quality coefficient at another random position is 0.8, which is significantly different from the current data quality coefficient, the cosine similarity is low and the matching degree is correspondingly low.

[0050] The inspection data optimization module 14 is used to optimize the selection of the inspection data generation network according to the data quality coefficient and the inspection position deviation coefficient, obtain an optimized generation network group, input the inspection data, obtain generated inspection data, and perform relay transmission, wherein the optimization adjustment direction is set according to the data quality coefficient.

[0051] In this embodiment, the optimization direction is determined based on the data quality coefficient. When the data quality coefficient is positive, it indicates that the actual data is more disturbed than predicted, and the optimization direction should focus on enhancing the anti-interference capability, selecting a generative network model with stronger noise suppression effect; if the data quality coefficient is negative, it indicates that the actual data quality is better than predicted, and the anti-interference processing intensity can be appropriately reduced, prioritizing the selection of a generative network that better preserves data details; when the data quality coefficient is close to zero, the optimization direction focuses on maintaining the stability of the current data quality, selecting a generative network with balanced overall performance.

[0052] Furthermore, generative adversarial networks can generate original inspection data based on the inspection data affected by electromagnetic interference, thereby improving the quality of inspection data and reducing electromagnetic interference.

[0053] Specifically, the inspection data optimization module 14 in the system includes: Obtain the optimized generation network group, which includes the total number of optimized generation networks; According to a preset number, several first generator network groups are randomly selected from the optimized generator network group, wherein the preset number is less than the total number; Randomly select training input data from several first generator network groups, analyze the similarity with the inspection data, and use it as multiple first selection fitness values; Based on the sign of the data quality coefficient, set the optimization direction; based on the absolute value of the data quality coefficient and the inspection position deviation coefficient, set the adjustment ratio. According to the optimization direction and adjustment ratio, based on multiple first selection fitness, several first generation network groups are adjusted and updated to obtain several first updated generation network groups; Continue to randomly select and update network groups, perform iterative optimization until convergence, obtain the optimized network group with the highest fitness, input the inspection data to obtain the generated inspection dataset, and perform relay transmission.

[0054] In this embodiment, firstly, based on a generative adversarial network (GAN) architecture, an optimized group of generative networks with various structures and parameter configurations is constructed. These networks differ in the number of network layers in the generator and discriminator, the number of neurons in the hidden layers, the type of activation function, the optimizer parameters, and the combination of loss functions, to cover different levels of anti-interference capabilities and data detail preservation characteristics. For example, some networks may use deeper convolutional layers to enhance the extraction and suppression of complex noise, while others may introduce attention mechanisms to focus on preserving detailed information of key feature regions of the device. The total number is determined based on the actual application scenario and computing resources, for example, set to 20 generative networks with different configurations.

[0055] Secondly, according to a preset number, such as 5, generator networks are randomly selected from the optimized generator network group and combined to form several first generator network groups. The preset number must be less than the total number of optimized generator networks to ensure sufficient combination diversity. Each first generator network group consists of multiple different generator networks to improve data optimization through ensemble learning.

[0056] Next, training input data is randomly selected for each first generator network group. The training input data comes from a dataset of electromagnetic interference environments similar to the current inspection scenario in historical inspection data, including raw inspection data with different noise levels and different equipment types, as well as their corresponding high-quality labeled data, or interference-free data obtained through other high-precision equipment.

[0057] Then, the similarity between the selected training input data and the current inspection data to be optimized is calculated. The similarity calculation can take into account the statistical characteristics of the data, such as the mean, variance, spectral characteristics and the degree of matching of key feature points. The calculation result is used as the first fitness of the first generation network group. The higher the fitness value, the better the potential optimization effect of the network group under the current data characteristics may be.

[0058] Then, the optimization direction is determined based on the sign of the data quality coefficient, and the adjustment ratio is set based on the absolute value of the data quality coefficient and the inspection position deviation coefficient. For example, the larger the absolute value of the data quality coefficient, the more significant the deviation between the actual data quality and the prediction, and the higher the adjustment ratio should be; the larger the inspection position deviation coefficient, the more likely the positioning error will introduce additional environmental interference uncertainty, and the adjustment ratio should be increased accordingly.

[0059] Subsequently, according to the set optimization direction and adjustment ratio, several first-selection network groups are adjusted and updated based on multiple first-choice fitness values. Specifically, the relatively weaker performing networks are replaced or adjusted according to the optimization direction, that is, the generated networks are adjusted according to the calculated adjustment ratio. For example, if the adjustment ratio is 30% and the total number of networks is 10, then 3 networks are adjusted according to the adjustment direction to approach the optimal network combination or move away from the worst network combination.

[0060] Finally, the process of randomly selecting network groups, calculating the fitness of the selected network, and adjusting and updating the network is repeated for iterative optimization. In each iteration, a new random combination and adjustment strategy are introduced to avoid getting trapped in local optima. The iterative process continues until the improvement in fitness of the selected network is less than a preset threshold (e.g., 0.01) in several consecutive iterations, or the maximum number of iterations (e.g., 100) is reached, at which point the optimization is considered to have converged.

[0061] At this point, the network with the highest fitness among all generator network groups is selected as the final optimized generator network group. The currently acquired inspection data is input into this optimized generator network group, and each generator network processes the data in parallel and outputs optimized data. Then, the output results of each network are merged through an integration strategy, such as weighted averaging or a voting mechanism, to obtain the final generated inspection dataset. This dataset has a higher signal-to-noise ratio and more complete detailed information. Subsequently, it is relayed to ensure that the backend monitoring center receives high-quality inspection data.

[0062] Among these, obtaining the optimized generation network group includes: Based on the inspection data transmission record, collect the original inspection data set of the samples, and collect the sample inspection data set after transmission; Based on generative adversarial networks, multiple optimized generative networks are constructed, each of which includes a generator and a discriminator; The sample inspection data set and the original sample inspection data set are divided multiple times to obtain multiple training data sets. The sample inspection data is used as input and the original sample inspection data is used as the discrimination label. The multiple optimized generative networks are iteratively trained until the test is passed, and an optimized generative network group is obtained.

[0063] In this embodiment of the application, firstly, sample data covering different electromagnetic environments, different equipment types, different weather conditions, and different inspection periods are selected from historical power inspection data transmission records.

[0064] Specifically, the raw inspection data collected and successfully transmitted by the UAV in an environment without significant electromagnetic interference is used as a sample raw inspection data set. The data in this set is characterized by high definition and low noise, and can serve as ideal target data. At the same time, inspection data actually transmitted to the monitoring center from the same or similar inspection locations and with the same equipment targets, but subjected to different degrees of electromagnetic interference, is also collected as a sample inspection data set. This set includes image quality degradation caused by electromagnetic interference.

[0065] Secondly, based on the basic architecture of generative adversarial networks (GANs), several optimized generative networks with differentiated structures and parameters are constructed. Each optimized generative network consists of a generator and a discriminator. The generator's function is to receive degraded sample inspection data as input, and through feature extraction, noise suppression, and detail restoration operations, output generated data that is as close as possible to the original sample inspection data.

[0066] Furthermore, its network structure adopts an improved architecture based on U-Net. For example, in the encoder part, convolutional layers and pooling layers are used to progressively downsample and extract multi-scale features, while in the decoder part, upsampling and skip connections are used to fuse feature information from different levels of the encoder. The generator can also introduce residual blocks to alleviate the gradient vanishing problem during deep network training, or introduce an attention gating mechanism to focus on the recovery of key areas of the device. The discriminator is responsible for judging the authenticity of the input image, i.e., the real sample raw inspection data or the generated data output by the generator. Its network structure can adopt a convolutional neural network, which extracts high-level semantic features of the image through layer-by-layer convolution and activation functions, such as LeakyReLU, and outputs a probability value representing the confidence that the input image is real raw data.

[0067] Then, the collected sample inspection data set and the original sample inspection data set are preprocessed, including data cleaning and standardization, such as normalizing pixel values ​​to the [0,1] range and data augmentation. Afterwards, according to a preset division ratio of 7:3, the preprocessed sample data is randomly divided into multiple training and test sets. For example, a total of 5 different training sets and corresponding test sets are divided to perform cross-validation, improving the stability and reliability of model training.

[0068] Next, for each constructed generative network, iterative training is performed using multiple pre-defined training datasets. During training, sample inspection data, i.e., degraded data, is input into the generator, which outputs preliminary generated data. The discriminator receives both the generated data and the corresponding original sample inspection data, and calculates the discriminative loss for both. The generator's loss function typically includes adversarial loss and content loss.

[0069] The adversarial loss, calculated from the discriminator's output, prompts the generator to produce data closer to reality. Content losses, such as mean squared error (MSE) and structural similarity loss (SSIM), measure the differences between generated and real data at the pixel or structural levels. By alternately training the generator and discriminator and continuously adjusting the network parameters, the data generated by the generator can increasingly realistically mimic the original inspection data, while the discriminator becomes increasingly unable to distinguish between generated and real data.

[0070] After each training round, the performance of the generator network is evaluated using the corresponding test set. Evaluation metrics may include Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and subjective visual effect score. Training stops when the evaluation metrics on the test set meet preset passing standards, such as a PSNR greater than 30 dB or an SSIM greater than 0.9, or after a preset maximum number of training rounds, such as 200 rounds. The above training and testing process is performed on all constructed generator networks. The generator networks that pass the test are then aggregated to form an optimized generator network group for subsequent inspection data optimization.

[0071] Furthermore, based on the sign of the data quality coefficient, an optimization direction is set; and based on the absolute value of the data quality coefficient and the inspection position deviation coefficient, an adjustment ratio is set, including: Obtain the sign of the data quality coefficient. If it is positive, the optimization direction is to move away from the generator network group with the smallest fitness. If it is negative, the optimization direction is to move closer to the generator network group with the largest fitness. Calculate the absolute value of the data quality coefficient and the average value of the inspection position deviation coefficient, and use them as the adjustment ratio.

[0072] In this embodiment, when determining the optimization direction, if the data quality coefficient is positive, it indicates that the current inspection data is subject to interference beyond expectations. In this case, it is necessary to strengthen the anti-interference capability. The optimization direction is set to move away from the generator network group with the lowest fitness, i.e., to reselect an optimized generator network with adjusted proportions, which is different from selecting the generator network group with the lowest fitness. This is because selecting the network group with the lowest fitness results in the worst anti-interference performance or data optimization effect when the current data quality coefficient is positive. Moving away from this network group means avoiding its inefficient network configuration and parameter combinations, thus moving closer to a better anti-interference network structure. For example, if a certain first generator network group has a much lower signal-to-noise ratio than other network groups when facing highly interfered data, and thus selects the network group with the lowest fitness, then in subsequent adjustment and update processes, the composition of this network group is adjusted to reduce the proportion of network structures similar to this network group, in order to move away from its disadvantageous characteristics.

[0073] Furthermore, if the data quality coefficient is negative, it indicates that the actual data quality is better than the prediction. The optimization direction is then set to approach the generator network group with the highest fitness, i.e., the optimized generator network with the adjusted update ratio is the same as the generator network group with the highest fitness. Choosing the generator network group with the highest fitness, given the current good data quality, can more effectively preserve data details or maintain the stability of high-quality data. Approaching this network group means borrowing its advantageous network architecture and parameter settings to further amplify its performance in preserving data details. For example, when a certain first generator network group processes low-noise inspection data, it outputs the highest structural similarity index, thus choosing the network group with the highest fitness. Subsequent adjustments are based on this network group, fine-tuning the structure of other network groups to make them closer to the characteristics of this optimal network group, in order to better adapt to scenarios where the data quality is better than the prediction.

[0074] Specifically, when setting the adjustment ratio, first calculate the absolute value of the data quality coefficient and the average value of the inspection position deviation coefficient. Add the absolute value of the data quality coefficient to the inspection position deviation coefficient and divide by 2; the result is the adjustment ratio.

[0075] For example, if the absolute value of the current data quality coefficient is 0.4 and the inspection position deviation coefficient is 0.6, then the average of the two is (0.4 + 0.6) / 2 = 0.5, and the adjustment ratio is 0.5. Therefore, 5 out of the 10 optimized generation networks will be selected for updating. This adjustment ratio reflects the required adjustment range of the generation network group under the combined effect of data quality deviation and position deviation. The higher the ratio, the greater the magnitude of structural adjustments, parameter updates, or member replacements to the network group.

[0076] In summary, compared with existing technologies, this application achieves adaptive optimization of inspection data by constructing an optimized generation network group containing various differentiated structures and parameter configurations, and dynamically adjusting the optimization direction and adjustment ratio of the network group based on the data quality coefficient and the inspection position deviation coefficient.

[0077] In summary, the embodiments of this application have at least the following technical effects: This application provides a dedicated UAV patrol relay data transmission system for power line inspection. First, it acquires the inspection location and data transmitted by the UAV and retrieves the corresponding power characteristics, providing foundational data for subsequent data quality assessment and network optimization. Second, it predicts the impact of location power characteristics on data quality and, combined with actual inspection data processing, obtains the actual impact coefficient. The difference between the two is calculated to obtain the data quality coefficient, providing a quantitative indicator for judging data reliability. Third, it analyzes the deviation of the inspection location based on the data quality coefficient. By randomly selecting other inspection locations and calculating their matching degree with the current location, it ultimately determines the inspection location deviation coefficient, identifying potential data acquisition problems due to location factors. Finally, it selects and optimizes the inspection data generation network. Through a preset number of random selections, similarity analysis, fitness assessment, and optimization direction and adjustment ratio settings based on the sign and absolute value of the data quality coefficient, iterative optimization is performed to obtain the optimal generation network group. The original inspection data is input into this network group to generate high-quality inspection data for relay transmission. This dynamic network selection and optimization adjustment mechanism can adapt to different transmission network environments based on real-time data quality and location deviation.

[0078] Through the above technical solution, this application improves the reliability of inspection data during transmission, ensures the accuracy of analysis and decision-making based on inspection data, and thus guarantees the efficient implementation of power inspection work.

[0079] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0080] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0081] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A dedicated UAV patrol relay data transmission system for power line inspection, characterized in that, The system includes: The inspection information acquisition module is used to acquire the inspection location and inspection data transmitted by the drone during power inspection, and retrieve the location power characteristics based on the inspection location. The quality impact prediction module is used to predict the impact of inspection data quality based on the location power characteristics, obtain the predicted data quality impact coefficient, obtain the actual data quality impact coefficient based on the inspection data processing, and calculate the data quality coefficient. The deviation coefficient acquisition module is used to analyze and obtain the inspection position deviation coefficient based on the data quality coefficient. The inspection data optimization module is used to optimize the selection of the inspection data generation network according to the data quality coefficient and the inspection position deviation coefficient, obtain an optimized generation network group, input the inspection data, obtain generated inspection data, and perform relay transmission. The optimization adjustment direction is set according to the data quality coefficient.

2. The power line inspection drone patrol relay data transmission system according to claim 1, characterized in that, Acquire the inspection location and inspection data transmitted by the drone during power line inspection, and retrieve the location power characteristics based on the inspection location, including: During the process of drone power line inspection, the inspection data transmitted by the drone and the drone's current inspection location are acquired. Based on the inspection location, the corresponding location power characteristics within the power network are obtained through indexing.

3. The power line inspection drone patrol relay data transmission system according to claim 2, characterized in that, Based on the inspection location, the corresponding location power characteristics within the power network are obtained through indexing, including: Based on the sensors in the power network, the distribution of monitored power characteristics within the inspection area is obtained; Based on the inspection location, the location power features are extracted within the power feature distribution.

4. The power line inspection dedicated UAV patrol relay data transmission system according to claim 1, characterized in that, Based on the location power characteristics, the impact of inspection data quality is predicted to obtain the predicted data quality impact coefficient. The actual data quality impact coefficient is obtained through inspection data processing. Finally, the data quality coefficient is calculated, including: The location power characteristics are input into the data quality impact predictor, and the predicted data quality impact coefficient is output. Based on the inspection data, the actual data quality impact coefficient is obtained, wherein the actual data quality impact coefficient includes the proportion of noisy data points in the inspection data; The difference between the actual data quality impact coefficient and the predicted data quality impact coefficient is calculated to obtain the data quality coefficient.

5. The power line inspection dedicated UAV patrol relay data transmission system according to claim 1, characterized in that, The training steps for the data quality impact predictor include: Based on the test records of the electromagnetic environment affecting the transmission data in the power area, a set of sample power characteristics was collected, and the data quality impact coefficients under different sample power characteristics were collected, labeled, and a set of sample predicted data quality impact coefficients was obtained. Based on machine learning, construct the structure of a data quality impact predictor; The data quality impact predictor is trained and tested under supervised supervision using the sample power feature set and the sample predicted data quality impact coefficient set as input data and labels until the test converges, thus completing the training.

6. The power line inspection dedicated UAV patrol relay data transmission system according to claim 1, characterized in that, Based on the data quality coefficient, the inspection location deviation coefficient is obtained through analysis, including: Randomly select other inspection locations within the inspection area and retrieve the power characteristics of the random locations; The first random location matching degree is obtained by analyzing the power characteristics of random locations and the data quality coefficient. Continue to randomly select other inspection locations to obtain the matching inspection location with the highest location matching degree, calculate the distance to the inspection location, and combine it with the distance scale of the inspection area to calculate the inspection location deviation coefficient.

7. The power line inspection drone patrol relay data transmission system according to claim 6, characterized in that, Based on the power characteristics of random locations and data quality coefficients, the first random location matching degree is obtained through analysis, including: The random location power characteristics are input into the data quality impact predictor, which outputs the random predicted data quality impact coefficient. Combined with the actual data quality impact coefficient, the first random data quality coefficient is calculated. Calculate the similarity between the first random data quality coefficient and the data quality coefficient, and use it as the first random location matching degree.

8. The power line inspection drone patrol relay data transmission system according to claim 1, characterized in that, Based on the data quality coefficient and inspection location deviation coefficient, the inspection data generation network is selected and optimized to obtain an optimized generation network group. The inspection data is then input to generate inspection data, which is then relayed, including: Obtain the optimized generation network group, which includes the total number of optimized generation networks; According to a preset number, several first generator network groups are randomly selected from the optimized generator network group, wherein the preset number is less than the total number; Randomly select training input data from several first generator network groups, analyze the similarity with the inspection data, and use it as multiple first selection fitness values; Based on the sign of the data quality coefficient, set the optimization direction; based on the absolute value of the data quality coefficient and the inspection position deviation coefficient, set the adjustment ratio. According to the optimization direction and adjustment ratio, based on multiple first selection fitness, several first generation network groups are adjusted and updated to obtain several first updated generation network groups; Continue to randomly select and update network groups, perform iterative optimization until convergence, obtain the optimized network group with the highest fitness, input the inspection data to obtain the generated inspection dataset, and perform relay transmission.

9. The power line inspection drone patrol relay data transmission system according to claim 8, characterized in that, Obtain optimized generator network groups, including: Based on the inspection data transmission record, collect the original inspection data set of the samples, and collect the sample inspection data set after transmission; Based on generative adversarial networks, multiple optimized generative networks are constructed, each of which includes a generator and a discriminator; The sample inspection data set and the original sample inspection data set are divided multiple times to obtain multiple training data sets. The sample inspection data is used as input and the original sample inspection data is used as the discrimination label. The multiple optimized generative networks are iteratively trained until the test is passed, and an optimized generative network group is obtained.

10. The power line inspection dedicated UAV patrol relay data transmission system according to claim 8, characterized in that, Based on the sign of the data quality coefficient, the optimization direction is set; based on the absolute value of the data quality coefficient and the inspection position deviation coefficient, the adjustment ratio is set, including: Obtain the sign of the data quality coefficient. If it is positive, the optimization direction is to move away from the generator network group with the smallest fitness. If it is negative, the optimization direction is to move closer to the generator network group with the largest fitness. Calculate the absolute value of the data quality coefficient and the average value of the inspection position deviation coefficient, and use them as the adjustment ratio.

Images