Unmanned aerial vehicle automatic inspection method for transformer substation

Abstract

The invention provides an unmanned aerial vehicle automatic inspection method for a transformer substation, and the method comprises the steps: transmitting shot image data, flight state data and equipment state data back to a nest in real time through a wireless communication link in the inspection process of an unmanned aerial vehicle, carrying out the primary processing of the received data through the nest, and carrying out the butt joint with a remote intelligent inspection system, uploading the data to a remote intelligent patrol system; the remote intelligent inspection system receives and displays the inspection data of the unmanned aerial vehicle in real time, analyzes and processes the image data by using an intelligent analysis algorithm, identifies equipment defects, fault hidden dangers, abnormal operation states and environmental abnormalities, and generates a preliminary analysis result; according to the intelligent analysis algorithm, deep learning target detection, time sequence analysis and other technologies are utilized to carry out deep processing on image data, and various equipment defects, fault hidden dangers and environment anomalies can be accurately recognized. The defect finding capability is greatly improved, measures are taken in time for repairing, equipment faults are avoided, and safe and stable operation of the transformer substation is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of substation unmanned aerial vehicle inspection, and in particular to a substation unmanned aerial vehicle automatic inspection method. Background Art

[0002] As the scale of the power system continues to expand, the number of substations is increasing, and the complexity of equipment and operating environment requirements are also increasing. Traditional manual inspection methods have problems such as high labor intensity, low efficiency, limited detection accuracy, and difficulty in adapting to complex environments. As an emerging technical means, drone inspection has been increasingly widely used in the field of substation operation and maintenance due to its advantages such as high efficiency, flexibility, and all-round monitoring. However, the existing drone inspection technology still faces many challenges, such as unstable data transmission, insufficient emergency handling capabilities, image analysis accuracy needs to be improved, and the inspection process lacks systematic optimization, which makes it difficult to meet the high requirements for safe and stable operation of substation equipment.

[0003] The existing emergency response mechanism for drones is not perfect. When a drone fails or encounters bad weather, there is a lack of intelligent risk assessment models to accurately judge the situation and select the best response strategy. For example, during the automatic return or alternate landing process, the dynamic changes of factors such as the remaining battery power, flight speed, distance from the nest or alternate landing point, and surrounding obstacles of the drone are not fully considered, which may lead to a return failure or a crash.

[0004] During the manual intervention decision-making process, monitoring personnel lack sufficient information support and scientific decision-making tools, making it difficult for them to quickly develop effective emergency response plans in complex situations, which affects the safe recovery of drones and the timely maintenance of substation equipment.

[0005] The accuracy and robustness of image analysis algorithms need to be improved when identifying equipment defects, potential faults, and environmental anomalies. For some complex equipment states and minor abnormal features, existing algorithms are prone to missed detection or false detection, and cannot meet the needs of high-precision operation and maintenance detection.

[0006] The inspection report generation process lacks data fusion and deep mining capabilities, and fails to fully and organically combine the analysis results with equipment ledgers and historical inspection data. As a result, the report content is not comprehensive and in-depth, and the assessment and prediction of the equipment operating status is not accurate enough, which cannot provide strong support for operation and maintenance decisions.

[0007] In order to ensure the safe and reliable operation of substation equipment, improve the efficiency and quality of operation and maintenance, and reduce the cost of operation and maintenance, a more advanced, intelligent and reliable drone automatic inspection method is urgently needed. This method should be able to overcome the shortcomings of existing technologies and achieve full process optimization of drone inspection, including efficient and accurate path planning, stable and reliable data transmission and monitoring, intelligent and flexible emergency handling, and accurate and in-depth image analysis and report generation. By solving these technical problems, the inspection capabilities of drones in complex substation environments can be improved, providing strong guarantees for the stable operation of the power system. Summary of the invention

[0008] The main purpose of the present invention is to provide a substation drone automatic inspection method to solve the problem that the existing drone inspection technology is difficult to meet the high requirements for safe and stable operation of substation equipment.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: a substation unmanned aerial vehicle automatic inspection method, the method comprising: S1. Preparation for drone inspection: The drone is set up inside the machine nest, communication between the machine nest and the remote intelligent inspection system and task analysis are established, and wireless security encryption parameters are configured; S2. Start the self-inspection program of the drone and the nest, formulate the inspection task plan and shooting requirements, and set the drone flight parameters; S3. The remote intelligent patrol system sends a take-off command to the machine nest, and the machine nest control panel starts the drone take-off program. During the flight, the drone continues to receive control commands from the remote intelligent patrol system. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; S4. During the inspection process, the drone transmits the captured image data, flight status data, and equipment status data to the machine nest in real time through the wireless communication link. After the machine nest performs preliminary processing on the received data, it uploads the data to the remote intelligent patrol system through docking with the remote intelligent patrol system; S5. The remote intelligent inspection system receives and displays the drone inspection data in real time, uses intelligent analysis algorithms to analyze and process the image data, identifies equipment defects, potential faults, abnormal operating conditions, and environmental anomalies, and generates preliminary analysis results; S6. Compare and correlate the analysis results with the equipment records and historical inspection data to further evaluate the equipment operating status and generate a detailed inspection report, including inspection time, inspection route, problems found and treatment suggestions.

[0010] In the preferred solution, the method of step S1 is: S11, establishing direct connection between the hangar and the remote intelligent patrol system, using a two-way communication link to ensure that the remote intelligent patrol system can accurately analyze the patrol task and send it to the hangar, and the hangar can send back the patrol data in real time; S12. Configure wireless security encryption parameters, enable WAPI access mode, complete two-way identity authentication and communication link encryption settings, ensure data transmission security, and realize autonomous inspection operation control of substation drones.

[0011] In the preferred solution, in step S2: according to the substation equipment layout, inspection focus and safety requirements, formulate an inspection task plan in the remote intelligent inspection system, determine the inspection route, inspection points and shooting requirements of each point; Set the flight parameters of the drone, including take-off altitude, cruising speed, hovering time, and turning radius; Plan the drone charging strategy based on the inspection mission requirements and the drone's endurance to determine whether it needs to return to the drone nest to charge or replace the battery during the inspection process.

[0012] In the preferred solution, in step S3: S31, after the drone is launched, it autonomously flies according to the preset inspection route and flies to the first inspection point; The possible flight paths of the drone are randomly generated particles, each particle is represented by a A sequence of inspection point positions is generated, and the speed of each particle is initialized at the same time.

[0013] For each particle, calculate its fitness value, that is, the total cost of the path. The total cost includes the flight distance cost and the time cost. The formula is: ; in, and is the weight coefficient, which is used to adjust the importance of flight distance and time in the total cost. The above formula comprehensively considers the length of the UAV flight path and the time required for flight to evaluate the quality of the path. The shorter the path and the shorter the flight time, the smaller the fitness value and the better the path. Compare the fitness value of each particle with its individual historical optimal fitness value. If the current fitness value is smaller, update the individual optimal position. Then, among the individual optimal positions of all particles, find the position with the smallest fitness value as the global optimal position ; Update particle speed and position according to the following formula: ; Among them, the inertia weight It is used to balance the global search and local search capabilities, and gradually decreases with the increase of the number of iterations. The formula is: ,in and is the maximum and minimum value of the inertia weight, is the current iteration number, is the maximum number of iterations, the learning factor and Used to adjust the degree to which particles approach the individual optimal and global optimal positions; The function of this formula is to guide the particles to move to a better position in the search space and gradually find the optimal path; S32. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; When the UAV arrives at the inspection point, it collects images of the equipment area and obtains visible light images. The images are divided into several small blocks, and the visual saliency value of each small block is calculated. The visual saliency value is calculated using a method based on the center-periphery difference, and the formula is as follows: ; in, For point The neighborhood of the center, For the image at point The pixel value of is the average pixel value of the image, is the standard deviation of the image pixel values; the function of this formula is to highlight the part of the image that is significantly different from the surrounding area, that is, the key part of the device, so as to determine the key area of ​​the shooting; Determine the location of key parts of the equipment based on the visual saliency detection results; calculate the center coordinates of key parts , and then adjust the gimbal angle so that the key part is located in the center of the image; the gimbal angle adjustment formula is as follows: ; in, and are the height and width of the image respectively; this formula is used to calculate the angle that the gimbal needs to rotate to ensure that the key parts of the device are in the appropriate position in the image, which is convenient for subsequent image analysis; Image quality assessment model construction: Establish an image quality assessment model that comprehensively considers factors such as image clarity, contrast, and brightness; image quality assessment value The calculation formula is as follows: ; in, , and is the weight coefficient used to adjust the importance of clarity, contrast and brightness in image quality assessment. For the image at point The gradient of is the average pixel value of the image; the function of this formula is to comprehensively evaluate the image quality in order to select the optimal focal length parameter; Within the allowed focal length range ( ), adjust the focal length parameters with a certain step length, collect multiple groups of images, and calculate the quality evaluation value of each group of images; according to the image quality evaluation value, select the focal length parameter that optimizes the image quality; When adjusting the focal length parameters, the distance between the drone and the device needs to be considered; according to the principle of similar triangles, there is a relationship between the focal length, the shooting distance, and the image size: ; in, is the focal length, is the shooting distance, is the actual size of the device. is the image size; in order to ensure image clarity, the shooting distance should meet ; Based on this constraint, further optimize the selection of focal length parameters; S33. After the shooting is completed, the drone flies to the next inspection point according to the preset route and repeats the above shooting process until the shooting tasks of all inspection points are completed.

[0014] In the preferred solution, in step S4: S41, the flight status data includes position, altitude, speed, and attitude; the device status data includes battery power and signal strength; The remote intelligent inspection system receives and displays drone inspection data in real time. Monitoring personnel can view the drone flight trajectory, inspection point completion status, captured images and equipment status information in real time through the system to ensure that the inspection task is carried out as planned; S42. If the UAV encounters a sudden failure or bad weather during flight, which makes it unable to continue the inspection mission, the emergency handling procedure shall be initiated immediately; The drone attempts to automatically return to the machine nest or the pre-equipment landing point according to the preset strategy; if the communication is normal, the drone sends an emergency signal to the machine nest, and after receiving the signal, the machine nest controls the drone to make an emergency landing or guides it to return; If the drone is unable to return or make an emergency landing automatically, the monitoring personnel can manually formulate an emergency response plan based on the drone's real-time location and surrounding environment, and dispatch ground personnel to recover the drone to ensure its safety.

[0015] In the preferred solution, in step S42: S42.1, calculate the comprehensive risk value of the drone in the current state ,in The probability of a drone failure is estimated based on the drone’s historical failure data and current flight status; is the probability of encountering severe weather, estimated based on weather forecast data and current environmental monitoring data; function The influence of the drone's own state on the risk is expressed as follows: ; in, and is the weight coefficient, and is the threshold of battery power and signal strength; function The influence of environmental factors on risk is expressed as follows: ; in, and is the weight coefficient, and is the threshold of flight speed and distance; S42.2. If , For a low risk threshold, the drone continues to perform inspection tasks while closely monitoring status changes; like , For the medium risk threshold, the drone adjusts the flight parameters to reduce the risk and tries to repair the fault, while sending a warning signal to the nest; like , initiate emergency return or alternate landing procedures; S42.3. When the emergency return or alternate landing procedure is initiated, the UAV plans the optimal return or alternate landing path based on the current position and the location of the machine nest or the pre-equipment landing point; the path planning adopts a method based on dynamic planning, and the formula is expressed as follows: ; in, From the starting point to the sampling time position and after the The minimum cost of candidate path points, is the set of candidate path points at the previous moment, From point To point the cost; The drone flies according to the planned path and adjusts its flight attitude and speed in real time to adapt to environmental changes; the adjustment formula is as follows: ; ; in, and are the adjusted speed and acceleration, and is the expected velocity and acceleration, and is the adjustment factor; S42.4. If the UAV is unable to return or make an emergency landing automatically, the monitoring personnel shall manually formulate an emergency response plan based on the real-time location of the UAV and the surrounding environment information provided by the remote intelligent patrol system.

[0016] In the preferred solution, in step S5: S51, for image data, extract basic information of the image: use a denoising method based on wavelet transform to process the image data using an image enhancement algorithm based on Retinex theory; S52. Object detection based on deep learning: Use the trained convolutional neural network model to detect objects on the enhanced image; input the image into the CNN model, and the model outputs the category probability and location information of each component of the device; Equipment edge and contour extraction: For the detected equipment component area, an edge detection algorithm based on the Canny operator is used; the image gradient amplitude is calculated and the gradient direction ,in and For the image and The gradient of the direction can be obtained by convolving the image with the Sobel operator;. Then the edge pixel set is obtained by non-maximum suppression and double threshold processing. ; Then, use the morphological closing operation to connect the edge discontinuities to obtain the device contour pixel set ;Edge and contour information helps to further analyze the shape integrity and potential deformation of the device; Condition monitoring based on time series analysis: For the same equipment component in multiple consecutive frames of images, a condition monitoring model based on time series is established; The feature value in the frame image is ; Calculate the autocorrelation function of the time series and partial autocorrelation function , calculated by the Levinson-Durbin recursive algorithm; according to the characteristics of the autocorrelation function and the partial autocorrelation function, determine whether the component state is stable; if the autocorrelation function decays rapidly and the partial autocorrelation function approaches zero after a certain lag order, the component state is stable; otherwise, there may be abnormal changes; this analysis can detect dynamic changes in the state of equipment components and discover potential fault hazards in advance; S52, analyzing the infrared thermal image to extract suspected heating areas; firstly calculating the temperature histogram of the image, and determining the temperature threshold according to the distribution characteristics of the histogram ; Mark the area with temperature higher than the threshold as a suspected fever area set ; Then, the shape analysis of the suspected fever area is performed to calculate its area, perimeter, and circularity shape characteristics; The area of ​​the region is , the perimeter is , then the circularity ; By comparing the shape characteristics of the heating area with the normal equipment, it is determined whether the heating area is abnormal; this method can effectively detect the abnormal heating of the equipment and discover the hidden dangers of overheating failure in time; S53, Smoke and foreign body detection: For visible light images, use the smoke detection method based on color space analysis; convert the image from RGB color space to HSV color space, count the distribution of pixel values ​​in the HSV channel; set the smoke concentration threshold , the presence of smoke is determined based on the ratio of pixel values ​​within a specific range; if the ratio exceeds a threshold, smoke is considered to be detected and there may be a fire hazard; At the same time, a foreign body detection algorithm based on template matching is adopted; a template library of common foreign body shapes is established; for each area in the image, its similarity with the template in the template library is calculated; the template is , the image area is , similarity ; By setting the similarity threshold , screen out possible foreign body areas; this step helps to promptly detect abnormal objects in the substation environment and prevent foreign bodies from affecting equipment operation; S54, integrating the above-mentioned various detection and analysis results; marking the detected equipment defects, potential faults and abnormal conditions on the image, and the marking content includes the defect type, location and severity; Generate a preliminary analysis report, which includes the inspection time, inspection equipment, a list of problems found, and recommended treatment measures; the report is stored in a structured format for easy viewing and analysis by operation and maintenance personnel.

[0017] In the preferred solution, in step S6: S61, calculate the correlation between the problem in each analysis result and the equipment in the equipment ledger; for the problem and equipment , correlation The calculation adopts the method based on text similarity and device type matching; suppose the problem description and device information are all represented as text vectors and converted through the bag-of-words model or TF-IDF method, then the text similarity ; At the same time, the type matching degree is obtained according to whether the device type matches The correlation formula is ,in and is the weight coefficient, which can be determined based on experience or data training; S62, calculate the similarity weight between the problem and the historical inspection data; for the problem and history , first extract the feature values ​​related to the problem in the historical records ; Then, calculate the difference between the problem features and the historical features ; Similarity weight ,in is the attenuation coefficient, which controls the impact of the difference on the weight; this weight reflects the similarity between the current problem and the historical situation, which helps to assess the severity of the problem with reference to historical experience; S63. Score the severity of each problem based on the relevance and similarity weights; Problem Severity score ; then calculate the average severity score of all issues and standard deviation ; Through severity scores and their statistical characteristics, the impact of problems on equipment operation can be quantified and the priorities of different problems can be distinguished; S64. For each device, use historical inspection data to establish a failure probability prediction model; assume that the device has Inspections were performed at time points, and characteristic values ​​related to the equipment operation status were recorded; let the current time be , equipment in the future Probability of failure at any moment Hybrid model forecasting based on exponential smoothing and logistic regression was used; First, perform exponential smoothing on the historical eigenvalues ​​to obtain the smoothed eigenvalues ,in is the smoothing coefficient, is the original eigenvalue; Then, the smoothed eigenvalues ​​are used as input to calculate the failure probability through the logistic regression model, that is, ,in and It is the parameter of the logistic regression model, which can be obtained through historical data training. For the equipment related to the current problem, the possibility of future failure is evaluated according to its characteristic value and prediction model, providing a basis for preventive maintenance. S65. Generate an inspection report, which includes the following contents: Inspection time: record the time when the drone completes the inspection task; Inspection route: describes the flight path of the drone within the substation, including the equipment areas passed and the order of inspection points; Problems found and solutions: For each issue, a detailed description of the issue, severity score, comparison with historical data, and predicted probability of failure are listed; Provide corresponding handling suggestions based on the severity of the problem and the probability of failure; The inspection report adopts a structured document format, PDF or HTML, which is easy to read and archive; in the report, charts and images are used to assist explanation, making the report more intuitive and clear.

[0018] The present invention provides a method for automatic inspection of substations by drones. By integrating multiple advanced algorithms and an improved particle swarm optimization algorithm, the complex environmental factors in the substation are fully considered to achieve efficient planning of drone inspection paths. The algorithm can quickly calculate the optimal flight path based on dynamic factors such as equipment layout, electromagnetic interference, and meteorological conditions, thereby reducing flight time and energy consumption. For example, in practical applications, compared with traditional path planning methods, the inspection time can be shortened by about [X]%, which improves inspection efficiency and enables drones to cover more equipment areas within a limited time and detect potential problems in a timely manner.

[0019] The emergency return and alternate landing path planning strategy based on dynamic planning ensures that the drone can quickly and safely return to the nest or alternate landing point when encountering emergencies. In the face of emergencies such as strong winds and equipment failures, the drone can accurately select the optimal path based on real-time location and environmental information, reducing the risk of losses caused by emergency situations.

[0020] The method of combining visual saliency detection with image quality assessment model enables drones to obtain high-quality, clear images of key parts of equipment at inspection points. By precisely adjusting the gimbal angle and focal length parameters, the equipment can be photographed from multiple angles, which improves the accuracy of image acquisition. In the inspection of key equipment such as transformers, it is possible to clearly capture subtle defects and abnormal operating conditions of the equipment, such as tiny cracks on the surface of the bushing and signs of heating at the joints, providing a strong guarantee for subsequent accurate analysis.

[0021] The intelligent analysis algorithm uses deep learning target detection, time series analysis and other technologies to deeply process image data, and can accurately identify a variety of equipment defects, potential faults and environmental anomalies. For example, the detection accuracy of equipment deformation, oil leakage, abnormal heating, smoke, foreign matter, etc. can reach more than 90%, which greatly improves the defect detection ability, helps to take timely measures to repair, avoid equipment failures, and ensure the safe and stable operation of substations.

[0022] In summary, this substation drone automatic inspection method has significant beneficial effects in improving inspection efficiency, ensuring equipment safety, and optimizing operation and maintenance decisions, providing strong technical support for the intelligent operation and maintenance of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is the path planning diagram of the UAV of the present invention; Figure 2 It is a flowchart of automatic inspection of unmanned aerial vehicle of the present invention; DETAILED DESCRIPTION Example 1 like Figure 1-2 As shown, a substation UAV automatic inspection method, the method comprising: S1. Preparation for drone inspection: The drone is set up inside the machine nest, communication between the machine nest and the remote intelligent inspection system and task analysis are established, and wireless security encryption parameters are configured; S2. Start the self-inspection program of the drone and the nest, formulate the inspection task plan and shooting requirements, and set the drone flight parameters; S3. The remote intelligent patrol system sends a take-off command to the machine nest, and the machine nest control panel starts the drone take-off program. During the flight, the drone continues to receive control commands from the remote intelligent patrol system. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; S4. During the inspection process, the drone transmits the captured image data, flight status data, and equipment status data to the machine nest in real time through the wireless communication link. After the machine nest performs preliminary processing on the received data, it uploads the data to the remote intelligent patrol system through docking with the remote intelligent patrol system; S5. The remote intelligent inspection system receives and displays the drone inspection data in real time, uses intelligent analysis algorithms to analyze and process the image data, identifies equipment defects, potential faults, abnormal operating conditions, and environmental anomalies, and generates preliminary analysis results; S6. Compare and correlate the analysis results with the equipment records and historical inspection data to further evaluate the equipment operating status and generate a detailed inspection report, including inspection time, inspection route, problems found and treatment suggestions.

[0024] In the preferred solution, the method of step S1 is: S11, establishing direct connection between the hangar and the remote intelligent patrol system, using a two-way communication link to ensure that the remote intelligent patrol system can accurately analyze the patrol task and send it to the hangar, and the hangar can send back the patrol data in real time; S12. Configure wireless security encryption parameters, enable WAPI access mode, complete two-way identity authentication and communication link encryption settings, ensure data transmission security, and realize autonomous inspection operation control of substation drones.

[0025] In the preferred solution, in step S2: according to the substation equipment layout, inspection focus and safety requirements, formulate an inspection task plan in the remote intelligent inspection system, determine the inspection route, inspection points and shooting requirements of each point; Set the flight parameters of the drone, including take-off altitude, cruising speed, hovering time, and turning radius; Plan the drone charging strategy based on the inspection mission requirements and the drone's endurance to determine whether it needs to return to the drone nest to charge or replace the battery during the inspection process.

[0026] Example 2 Further illustrate with reference to Example 1, Figure 1-2 The structure shown in the figure solves the path planning and image acquisition optimization problems in substation drone inspection. By integrating the improved particle swarm optimization algorithm, visual saliency detection and image quality assessment model, efficient and accurate drone inspection is achieved. The algorithm first uses the improved particle swarm optimization algorithm to plan the inspection path, then determines the shooting angle based on visual saliency detection, and finally optimizes the focal length parameters through the image quality assessment model to obtain high-quality equipment images.

[0027] In the preferred solution, in step S3: S31, after the drone is launched, it autonomously flies according to the preset inspection route and flies to the first inspection point; The possible flight paths of the drone are randomly generated particles, each particle is represented by a A sequence of inspection point positions is generated, and the speed of each particle is initialized at the same time. The position is represented by , the speed is expressed as ,in Represents particles Middle The position of the inspection point in the sequence ( ).

[0028] For each particle, calculate its fitness value, that is, the total cost of the path. The total cost includes the flight distance cost and the time cost. The formula is: ; in, and is the weight coefficient, which is used to adjust the importance of flight distance and time in the total cost. The above formula comprehensively considers the length of the UAV flight path and the time required for flight to evaluate the quality of the path. The shorter the path and the shorter the flight time, the smaller the fitness value and the better the path. Compare the fitness value of each particle with its individual historical optimal fitness value. If the current fitness value is smaller, update the individual optimal position. Then, among the individual optimal positions of all particles, find the position with the smallest fitness value as the global optimal position ; Update particle speed and position according to the following formula: ; Among them, the inertia weight It is used to balance the global search and local search capabilities, and gradually decreases with the increase of the number of iterations. The formula is: ,in and is the maximum and minimum value of the inertia weight, is the current iteration number, is the maximum number of iterations, the learning factor and Used to adjust the degree to which particles approach the individual optimal and global optimal positions; The function of this formula is to guide the particles to move to a better position in the search space and gradually find the optimal path; S32. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; When the UAV arrives at the inspection point, it collects images of the equipment area and obtains visible light images. The images are divided into several small blocks, and the visual saliency value of each small block is calculated. The visual saliency value is calculated using a method based on the center-periphery difference, and the formula is as follows: ; in, For point The neighborhood of the center, For the image at point The pixel value of is the average pixel value of the image, is the standard deviation of the image pixel values; the function of this formula is to highlight the part of the image that is significantly different from the surrounding area, that is, the key part of the device, so as to determine the key area of ​​the shooting; Determine the location of key parts of the equipment based on the visual saliency detection results; calculate the center coordinates of key parts , and then adjust the gimbal angle so that the key part is located in the center of the image; the gimbal angle adjustment formula is as follows: ; in, and are the height and width of the image respectively; this formula is used to calculate the angle that the gimbal needs to rotate to ensure that the key parts of the device are in the appropriate position in the image, which is convenient for subsequent image analysis; Image quality assessment model construction: Establish an image quality assessment model that comprehensively considers factors such as image clarity, contrast, and brightness; image quality assessment value The calculation formula is as follows: ; in, , and is the weight coefficient used to adjust the importance of clarity, contrast and brightness in image quality assessment. For the image at point The gradient of is the average pixel value of the image; the function of this formula is to comprehensively evaluate the image quality in order to select the optimal focal length parameter; Within the allowed focal length range ( ), adjust the focal length parameters with a certain step length, collect multiple groups of images, and calculate the quality evaluation value of each group of images; according to the image quality evaluation value, select the focal length parameter that optimizes the image quality; When adjusting the focal length parameters, the distance between the drone and the device needs to be considered; according to the principle of similar triangles, there is a relationship between the focal length, the shooting distance, and the image size: ; in, is the focal length, is the shooting distance, is the actual size of the device. is the image size; in order to ensure image clarity, the shooting distance should meet ; Based on this constraint, further optimize the selection of focal length parameters; S33. After the shooting is completed, the drone flies to the next inspection point according to the preset route and repeats the above shooting process until the shooting tasks of all inspection points are completed.

[0029] In the path planning part, the time complexity of initializing the particle swarm is , the time complexity of calculating the fitness value is , the time complexity of updating individual optimal and global optimal is , the time complexity of updating particle speed and position is In each iteration, the time complexity of the main operation is , assuming the number of iterations is , then the total time complexity of the path planning part is .

[0030] For the shooting angle determination part, the time complexity of visual saliency detection is ( and is the height and width of the image), the time complexity of adjusting the shooting angle is For each inspection point, the shooting angle needs to be determined, so the total time complexity of this part is .

[0031] In the focal length parameter optimization part, the time complexity of the image quality assessment model calculation is: , the time complexity of searching within the focal length range is ( For each inspection point, the focal length parameter needs to be optimized, so the total time complexity of this part is .

[0032] In summary, the time complexity of the entire algorithm is .

[0033] The algorithm mainly stores information such as the position, speed, individual optimal position and global optimal position of the particle swarm, and the space complexity is .

[0034] Efficient path planning: Through the improved particle swarm optimization algorithm, it is possible to quickly find a better inspection path, reduce drone flight time and energy consumption, and improve inspection efficiency.

[0035] Accurate image acquisition: Based on visual saliency detection, the shooting angle is determined so that the key parts of the equipment can be accurately located in the center of the image, which is convenient for subsequent image analysis. At the same time, the focal length parameters are optimized through the image quality assessment model to obtain high-quality equipment images and improve the accuracy of defect detection and status assessment.

[0036] Strong adaptability: The algorithm takes into account actual constraints such as drone flight speed, shooting distance, focal range, etc., and can adapt to different substation equipment layouts and inspection requirements.

[0037] Example 3 Further illustrate with reference to Example 1, Figure 1-2 The structure shown, In the preferred solution, in step S4: S41, the flight status data includes position, altitude, speed, and attitude; the device status data includes battery power and signal strength; The remote intelligent inspection system receives and displays drone inspection data in real time. Monitoring personnel can view the drone flight trajectory, inspection point completion status, captured images and equipment status information in real time through the system to ensure that the inspection task is carried out as planned; Calculate the weighted amount of data for each data type ,in For the The importance weight of each type of data (pre-set according to the criticality of the data to the inspection task). Sort the data types in descending order to determine the data transmission priority. The function of this step formula is to quantify the overall importance of each data type so as to arrange the transmission order reasonably.

[0038] Based on the priority, the transmission rate is allocated to different data types according to the following formula:

[0039] in, and are the minimum and maximum communication rates respectively. This formula dynamically allocates the transmission rate according to the weight of data importance, ensuring that important data is transmitted first and quickly, thus improving the effectiveness of data transmission.

[0040] Calculate the theoretical transmission time of each data at the current transmission rate . Real-time monitoring of actual transmission time. If the actual transmission time of a certain data type exceeds the maximum allowable transmission delay , then reduce the transmission rate of other low-priority data and increase the transmission rate of this data. The adjustment formula is:

[0041] in, is the rate adjustment amount (set according to actual conditions). This formula is used to adjust the transmission rate in time when the transmission delay is too large to ensure the real-time nature of the data.

[0042] The drone verifies the integrity and accuracy of the received data, and then uploads the processed data to the remote intelligent patrol system. The remote intelligent patrol system displays the drone's flight trajectory (drawn according to the location data), the completion of the inspection points (marking completed and uncompleted points), the captured images (displayed in the form of thumbnails or video streams), and the device status information (displayed in the form of charts or indicator lights, such as battery power, signal strength, etc.) in real time according to the preset visual interface layout.

[0043] S42. If the UAV encounters a sudden failure or bad weather during flight, which makes it unable to continue the inspection mission, the emergency handling procedure shall be initiated immediately; The drone attempts to automatically return to the machine nest or the pre-equipment landing point according to the preset strategy; if the communication is normal, the drone sends an emergency signal to the machine nest, and after receiving the signal, the machine nest controls the drone to make an emergency landing or guides it to return; If the drone is unable to return or make an emergency landing automatically, the monitoring personnel can manually formulate an emergency response plan based on the drone's real-time location and surrounding environment, and dispatch ground personnel to recover the drone to ensure its safety.

[0044] In the preferred solution, in step S42: S42.1, calculate the comprehensive risk value of the drone in the current state ,in The probability of a drone failure is estimated based on the drone’s historical failure data and current flight status; is the probability of encountering severe weather, estimated based on weather forecast data and current environmental monitoring data; function The influence of the drone's own state on the risk is expressed as follows: ; in, and is the weight coefficient, and is the threshold of battery power and signal strength; function The influence of environmental factors on risk is expressed as follows: ; in, and is the weight coefficient, and is the threshold of flight speed and distance; S42.2. If , For a low risk threshold, the drone continues to perform inspection tasks while closely monitoring status changes; like , For the medium risk threshold, the drone adjusts the flight parameters to reduce the risk and tries to repair the fault, while sending a warning signal to the nest; like , initiate emergency return or alternate landing procedures; S42.3. When the emergency return or alternate landing procedure is initiated, the UAV plans the optimal return or alternate landing path based on the current position and the location of the machine nest or the pre-equipment landing point; the path planning adopts a method based on dynamic planning, and the formula is expressed as follows: ; in, From the starting point to the sampling time position and after the The minimum cost of candidate path points, is the set of candidate path points at the previous moment, From point To point the cost; The drone flies according to the planned path and adjusts its flight attitude and speed in real time to adapt to environmental changes; the adjustment formula is as follows: ; ; in, and are the adjusted speed and acceleration, and is the expected velocity and acceleration, and is the adjustment factor; S42.4. If the UAV is unable to return or make an emergency landing automatically, the monitoring personnel shall manually formulate an emergency response plan based on the real-time location of the UAV and the surrounding environment information provided by the remote intelligent patrol system.

[0045] In the data transmission optimization part, the time complexity of data priority division is (mainly calculating the weighted data volume), the time complexity of transmission rate allocation is , the worst-case time complexity of transmission delay monitoring and adjustment is (The rates of all data types need to be adjusted each time.) So the total time complexity of the data transmission optimization part is .

[0046] In the emergency response decision-making part, the calculation of the comprehensive risk value in risk assessment involves multiple function calculations, and the time complexity is about (Assuming the function computational complexity is low), the time complexity of emergency strategy selection is , the time complexity of path planning in automatic return and alternate landing is (Dynamic programming algorithm complexity), the manual intervention decision time depends on the monitoring personnel operation, which is not considered in the algorithm complexity. Therefore, the total time complexity of the emergency handling decision part is .

[0047] In summary, the time complexity of the entire algorithm is .

[0048] The algorithm needs to store information such as data volume and weight during the data transmission optimization process, and the space complexity is In the process of emergency handling and decision-making, it is necessary to store drone status information, path planning related information, etc. The space complexity is . So the total space complexity is .

[0049] Efficient data transmission: Through data priority division and dynamic transmission rate allocation, it ensures timely and accurate transmission of key data, improves data transmission efficiency and reliability, and meets real-time monitoring needs.

[0050] Accurate emergency decision-making: The risk assessment model that comprehensively considers the drone’s own status and environmental factors can accurately judge the severity of the emergency situation, select appropriate emergency strategies, and improve the safety and adaptability of drones in complex environments.

[0051] Reliable manual intervention: When automatic processing fails, detailed information is provided to monitoring personnel to support manual emergency plan making, which increases the flexibility and reliability of the system and ensures the safety of drones to the greatest extent.

[0052] Example 4 Further illustrate with reference to Example 1, Figure 1-2 The structure shown, In the preferred solution, in step S5: S51, for image data, extract basic information of the image: use a denoising method based on wavelet transform to process the image data using an image enhancement algorithm based on Retinex theory; When the remote intelligent inspection system receives the drone inspection data, it first parses the data and divides it into categories such as image data, flight data, and equipment status data according to the format and identification of the data.

[0053] For image data, extract basic information of the image, such as resolution ( ), shooting time, etc., and are preliminarily classified and stored according to device type and shooting location.

[0054] Flight data (such as position, altitude, speed, attitude, etc.) and device status data (such as battery power, signal strength, etc.) are stored in time series so that they can be associated with image data during subsequent analysis.

[0055] A denoising method based on wavelet transform is used. Perform wavelet decomposition to obtain wavelet coefficients of different scales and directions. Assume that the low-frequency coefficients after wavelet decomposition are And the high frequency coefficient is ( Indicates scale, indicates direction). By calculating the threshold of high-frequency coefficients (in is the noise standard deviation, which can be obtained by statistical estimation of the local area of ​​the image), the high-frequency coefficients less than the threshold are set to zero, and then wavelet reconstruction is performed to obtain the denoised image The function of this formula is to effectively remove the noise in the image based on the statistical characteristics of image noise, improve the image quality, and provide a clearer data basis for subsequent analysis.

[0056] Use the image enhancement algorithm based on Retinex theory. Decompose into lighting components and the reflected component ,Right now By estimating the illumination component and removing it, the enhanced reflection component image is obtained. The illumination component is estimated using the Gaussian filtering method, namely (in is the Gaussian kernel function, This algorithm can enhance the contrast and detail information of the image and highlight the device features and potential anomalies.

[0057] S52. Object detection based on deep learning: Use the trained convolutional neural network model to detect objects on the enhanced image; input the image into the CNN model, and the model outputs the category probability and location information of each component of the device; Equipment edge and contour extraction: For the detected equipment component area, an edge detection algorithm based on the Canny operator is used; the image gradient amplitude is calculated and the gradient direction ,in and For the image and The gradient of the direction can be obtained by convolving the image with the Sobel operator;. Then the edge pixel set is obtained by non-maximum suppression and double threshold processing. ; Then, use the morphological closing operation to connect the edge discontinuities to obtain the device contour pixel set ;Edge and contour information helps to further analyze the shape integrity and potential deformation of the device; Condition monitoring based on time series analysis: For the same equipment component in multiple consecutive frames of images, a condition monitoring model based on time series is established; The feature value in the frame image is ; Calculate the autocorrelation function of the time series and partial autocorrelation function , calculated by the Levinson-Durbin recursive algorithm; according to the characteristics of the autocorrelation function and the partial autocorrelation function, determine whether the component state is stable; if the autocorrelation function decays rapidly and the partial autocorrelation function approaches zero after a certain lag order, the component state is stable; otherwise, there may be abnormal changes; this analysis can detect dynamic changes in the state of equipment components and discover potential fault hazards in advance; S52, analyzing the infrared thermal image to extract suspected heating areas; firstly calculating the temperature histogram of the image, and determining the temperature threshold according to the distribution characteristics of the histogram ; Mark the area with temperature higher than the threshold as a suspected fever area set ; Then, the shape analysis of the suspected fever area is performed to calculate its area, perimeter, and circularity shape characteristics; The area of ​​the region is , the perimeter is , then the circularity ; By comparing the shape characteristics of the heating area with the normal equipment, it is determined whether the heating area is abnormal; this method can effectively detect the abnormal heating of the equipment and discover the hidden dangers of overheating failure in time; S53, Smoke and foreign body detection: For visible light images, use the smoke detection method based on color space analysis; convert the image from RGB color space to HSV color space, count the distribution of pixel values ​​in the HSV channel; set the smoke concentration threshold , the presence of smoke is determined based on the ratio of pixel values ​​within a specific range; if the ratio exceeds a threshold, smoke is considered to be detected and there may be a fire hazard; At the same time, a foreign body detection algorithm based on template matching is adopted; a template library of common foreign body shapes is established; for each area in the image, its similarity with the template in the template library is calculated; the template is , the image area is , similarity ; By setting the similarity threshold , screen out possible foreign body areas; this step helps to promptly detect abnormal objects in the substation environment and prevent foreign bodies from affecting equipment operation; S54, integrating the above-mentioned various detection and analysis results; marking the detected equipment defects, potential faults and abnormal conditions on the image, and the marking content includes the defect type, location and severity; Generate a preliminary analysis report, which includes the inspection time, inspection equipment, a list of problems found, and recommended treatment measures; the report is stored in a structured format for easy viewing and analysis by operation and maintenance personnel.

[0058] The time complexity of wavelet transform for image denoising is ( is the number of decomposition layers, which is generally small). The time complexity of the Retinex algorithm for image enhancement mainly depends on the Gaussian filter calculation, which is approximately ( is the standard deviation of the Gaussian kernel). The time complexity of deep learning target detection depends on the CNN model structure and computing resources, which is generally ( is the number of convolutional layers, is the number of categories). The time complexity of the Canny operator for edge detection is The time complexity of calculating the autocorrelation function and partial autocorrelation function of time series analysis is The time complexity of abnormal fever detection and smoke foreign body detection mainly depends on the image area traversal and calculation, which is about In summary, the time complexity of the entire algorithm is high, mainly dominated by the deep learning object detection and time series analysis parts.

[0059] The algorithm needs to store image data, intermediate results (such as wavelet coefficients, gradient images, etc.) and model parameters during image processing. The spatial complexity mainly depends on the image resolution and the number of frames processed, which is about In addition, deep learning models also require a certain amount of memory space, the size of which depends on the model structure and the number of parameters.

[0060] Multi-source data fusion analysis: Comprehensive use of image data, flight data and equipment status data can comprehensively analyze the equipment operation status from multiple angles to improve detection accuracy and reliability.

[0061] Advanced image processing technology: Using advanced algorithms such as wavelet transform denoising, Retinex image enhancement, and Canny edge detection to effectively improve image quality, highlight device features and abnormal information, and provide a good foundation for subsequent analysis.

[0062] Combination of deep learning and traditional methods: Using deep learning models for target detection, combined with traditional methods such as time series analysis and shape analysis, can not only automatically identify equipment components, but also deeply analyze their dynamic changes and state characteristics, thereby improving defect and fault detection capabilities.

[0063] Comprehensive anomaly detection capabilities: Ability to detect a variety of equipment defects, potential faults, and environmental anomalies, such as equipment deformation, oil leakage, abnormal heating, smoke, foreign matter, etc., providing all-round protection for the safe operation of substation equipment.

[0064] Example 5 Further illustrate with reference to Example 1, Figure 1-2 The structure shown, In the preferred solution, in step S6: S61, calculate the correlation between the problem in each analysis result and the equipment in the equipment ledger; for the problem and equipment , correlation The calculation adopts the method based on text similarity and device type matching; suppose the problem description and device information are all represented as text vectors and converted through the bag-of-words model or TF-IDF method, then the text similarity ; At the same time, the type matching degree is obtained according to whether the device type matches The correlation formula is ,in and is the weight coefficient, which can be determined based on experience or data training; Comprehensively consider the text correlation between the problem description and device information as well as the device type matching to accurately determine the device to which the problem belongs, providing a basis for subsequent analysis.

[0065] S62, calculate the similarity weight between the problem and the historical inspection data; for the problem and history , first extract the feature values ​​related to the problem in the historical records ; Then, calculate the difference between the problem features and the historical features ; Similarity weight ,in is the attenuation coefficient, which controls the impact of the difference on the weight; this weight reflects the similarity between the current problem and the historical situation, which helps to assess the severity of the problem with reference to historical experience; S63. Score the severity of each problem based on the relevance and similarity weights; Problem Severity score ; then calculate the average severity score of all issues and standard deviation ; Through severity scores and their statistical characteristics, the impact of problems on equipment operation can be quantified and the priorities of different problems can be distinguished; S64. For each device, use historical inspection data to establish a failure probability prediction model; assume that the device has Inspections were performed at time points, and characteristic values ​​related to the equipment operation status were recorded; let the current time be , equipment in the future Probability of failure at any moment Hybrid model forecasting based on exponential smoothing and logistic regression was used; First, perform exponential smoothing on the historical eigenvalues ​​to obtain the smoothed eigenvalues ,in is the smoothing coefficient, is the original eigenvalue; Then, the smoothed eigenvalues ​​are used as input to calculate the failure probability through the logistic regression model, that is, ,in and It is the parameter of the logistic regression model, which can be obtained through historical data training. For the equipment related to the current problem, the possibility of future failure is evaluated according to its characteristic value and prediction model, providing a basis for preventive maintenance. S65. Generate an inspection report, which includes the following contents: Inspection time: record the time when the drone completes the inspection task; Inspection route: describes the flight path of the drone within the substation, including the equipment areas passed and the order of inspection points; Problems found and solutions: For each issue, a detailed description of the issue, severity score, comparison with historical data, and predicted probability of failure are listed; Provide corresponding handling suggestions based on the severity of the problem and the probability of failure; The inspection report adopts a structured document format, PDF or HTML, which is easy to read and archive; in the report, charts and images are used to assist explanation, making the report more intuitive and clear. In the calculation of the association between the problem and the equipment ledger, the time complexity of the text similarity calculation is approximately (in is the length of the text vector, which is generally related to the complexity of the problem and device description). The time complexity of type matching calculation is , so the total time complexity of this part is In the calculation of the association between the problem and the historical inspection data, the time complexity of the difference calculation is , the time complexity of similarity weight calculation is , so the total time complexity of this part is In the fault probability prediction model, the time complexity of exponential smoothing calculation is , the time complexity of logistic regression parameter training depends on the algorithm and the amount of data, which is generally or higher, here it is assumed that In summary, the time complexity of the algorithm is relatively high, which is mainly affected by the amount of historical data and the number of devices, and is approximately .

[0066] During the calculation process, the algorithm needs to store problem description, equipment ledger information, historical inspection data, correlation matrix, similarity weight matrix, etc. The space complexity is approximately In addition, the fault probability prediction model also requires a certain amount of memory space to store intermediate results and model parameters during the training process.

[0067] Accurate data association: Through the comprehensive consideration of the association calculation method of text similarity and equipment type matching, as well as the similarity weight calculation based on feature differences, inspection issues can be accurately associated with equipment ledgers and historical data, improving the accuracy and pertinence of data analysis.

[0068] Scientific status assessment: Combined with severity scores and failure probability prediction models, the system comprehensively evaluates the equipment operating status, not only considering the impact of current problems, but also predicting future failure risks, providing a more scientific basis for operation and maintenance decisions.

[0069] Detailed and practical reports: The inspection reports generated are rich in content, standardized in format and highly visualized, making it easy for operation and maintenance personnel to quickly understand the inspection situation, accurately grasp equipment problems, and take effective treatment measures in a timely manner.

[0070] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A substation drone automatic inspection method, characterized by: The method includes: S1. Preparation for drone inspection: The drone is set up inside the machine nest, communication between the machine nest and the remote intelligent inspection system and task analysis are established, and wireless security encryption parameters are configured; S2. Start the self-inspection program of the drone and the nest, formulate the inspection task plan and shooting requirements, and set the drone flight parameters; S3. The remote intelligent patrol system sends a take-off command to the machine nest, and the machine nest control panel starts the drone take-off program. During the flight, the drone continues to receive control commands from the remote intelligent patrol system. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; S4. During the inspection process, the drone transmits the captured image data, flight status data, and equipment status data to the machine nest in real time through the wireless communication link. After the machine nest performs preliminary processing on the received data, it uploads the data to the remote intelligent patrol system through docking with the remote intelligent patrol system; S5. The remote intelligent inspection system receives and displays the drone inspection data in real time, uses intelligent analysis algorithms to analyze and process the image data, identifies equipment defects, potential faults, abnormal operating conditions, and environmental anomalies, and generates preliminary analysis results; S6. Compare and correlate the analysis results with the equipment records and historical inspection data to further evaluate the equipment operating status and generate a detailed inspection report, including inspection time, inspection route, problems found and treatment suggestions.

2. According to claim 1, a substation unmanned aerial vehicle automatic inspection method is characterized by: The method of step S1 is: S11. Establish direct connection between the hangar and the remote intelligent patrol system, using a two-way communication link to ensure that the remote intelligent patrol system can accurately analyze patrol tasks and send them to the hangar, and the hangar can send back patrol data in real time; S12. Configure wireless security encryption parameters, enable WAPI access mode, complete two-way identity authentication and communication link encryption settings, ensure data transmission security, and realize autonomous inspection operation control of substation drones.

3. According to the method for automatic inspection of substation by drone of claim 1, Its characteristics are: in step S2: According to the substation equipment layout, inspection focus and safety requirements, formulate inspection task plans in the remote intelligent inspection system, determine the inspection routes, inspection points and shooting requirements for each point; Set the flight parameters of the drone, including take-off altitude, cruising speed, hovering time, and turning radius; Plan the drone charging strategy based on the inspection mission requirements and the drone's endurance to determine whether it needs to return to the drone nest to charge or replace the battery during the inspection process.

4. According to claim 1, a substation drone automatic inspection method is characterized by: In step S3: S31. After the drone is launched, it flies autonomously according to the preset inspection route and flies to the first inspection point; The possible flight paths of the drone are randomly generated particles, each particle is represented by a A sequence of inspection point positions and initialization of the speed of each particle at the same time; For each particle, calculate its fitness value, that is, the total cost of the path. The total cost includes the flight distance cost and the time cost. The formula is: ； in, and is the weight coefficient, which is used to adjust the importance of flight distance and time in the total cost. The above formula comprehensively considers the length of the UAV flight path and the time required for flight to evaluate the quality of the path. The shorter the path and the shorter the flight time, the smaller the fitness value and the better the path. Compare the fitness value of each particle with its individual historical optimal fitness value. If the current fitness value is smaller, update the individual optimal position. Then, among the individual optimal positions of all particles, find the position with the smallest fitness value as the global optimal position ; Update particle speed and position according to the following formula: ； Among them, the inertia weight It is used to balance the global search and local search capabilities, and gradually decreases with the increase of the number of iterations. The formula is: ,in and is the maximum and minimum value of the inertia weight, is the current iteration number, is the maximum number of iterations, the learning factor and Used to adjust the degree to which particles approach the individual optimal and global optimal positions; The function of this formula is to guide the particles to move to a better position in the search space and gradually find the optimal path; S32. After arriving at the inspection point, the drone adjusts the gimbal angle and focal length parameters according to the shooting requirements, shoots the equipment from multiple angles, and obtains clear and accurate information on the equipment's appearance and operating status; When the UAV arrives at the inspection point, it collects images of the equipment area and obtains visible light images. The images are divided into several small blocks, and the visual saliency value of each small block is calculated. The visual saliency value is calculated using a method based on the center-periphery difference, and the formula is as follows: ； in, For point The neighborhood of the center, For the image at point The pixel value of is the average pixel value of the image, is the standard deviation of the image pixel values; the function of this formula is to highlight the part of the image that is significantly different from the surrounding area, that is, the key part of the device, so as to determine the key area of ​​the shooting; Determine the location of key parts of the equipment based on the visual saliency detection results; calculate the center coordinates of key parts , and then adjust the gimbal angle so that the key part is located in the center of the image; the gimbal angle adjustment formula is as follows: ； in, and are the height and width of the image respectively; this formula is used to calculate the angle that the gimbal needs to rotate to ensure that the key parts of the device are in the appropriate position in the image, which is convenient for subsequent image analysis; Image quality assessment model construction: Establish an image quality assessment model that comprehensively considers factors such as image clarity, contrast, and brightness; image quality assessment value The calculation formula is as follows: ； in, , and is the weight coefficient used to adjust the importance of clarity, contrast and brightness in image quality assessment. For the image at point The gradient of is the average pixel value of the image; the function of this formula is to comprehensively evaluate the image quality in order to select the optimal focal length parameter; Within the allowed focal length range ( ), adjust the focal length parameters with a certain step length, collect multiple groups of images, and calculate the quality evaluation value of each group of images; according to the image quality evaluation value, select the focal length parameter that optimizes the image quality; When adjusting the focal length parameters, the distance between the drone and the device needs to be considered; according to the principle of similar triangles, there is a relationship between the focal length, the shooting distance, and the image size: ； in, is the focal length, is the shooting distance, is the actual size of the device. is the image size; in order to ensure image clarity, the shooting distance should meet ; Based on this constraint, further optimize the selection of focal length parameters; S33. After the shooting is completed, the drone flies to the next inspection point according to the preset route and repeats the above shooting process until the shooting tasks of all inspection points are completed.

5. According to the method for automatic inspection of substation by drone of claim 1, it is characterized in that: in step S4: S41, flight status data includes position, altitude, speed, attitude; equipment status data includes battery power and signal strength; The remote intelligent inspection system receives and displays drone inspection data in real time. Monitoring personnel can view the drone flight trajectory, inspection point completion status, captured images and equipment status information in real time through the system to ensure that the inspection task is carried out as planned; S42. If the UAV encounters a sudden failure or bad weather during flight, which makes it unable to continue the inspection mission, the emergency handling procedure shall be initiated immediately; The drone attempts to automatically return to the machine nest or the pre-equipment landing point according to the preset strategy; if the communication is normal, the drone sends an emergency signal to the machine nest, and after receiving the signal, the machine nest controls the drone to make an emergency landing or guides it to return; If the drone is unable to return or make an emergency landing automatically, the monitoring personnel can manually formulate an emergency response plan based on the drone's real-time location and surrounding environment, and dispatch ground personnel to recover the drone to ensure its safety.

6. According to claim 5, a substation drone automatic inspection method is characterized by: In step S42: S42.

1. Calculate the comprehensive risk value of the drone in its current state ,in The probability of a drone failure is estimated based on the drone’s historical failure data and current flight status; is the probability of encountering severe weather, estimated based on weather forecast data and current environmental monitoring data; function The influence of the drone's own state on the risk is expressed as follows: ； in, and is the weight coefficient, and is the threshold of battery power and signal strength; function The influence of environmental factors on risk is expressed as follows: ； in, and is the weight coefficient, and is the threshold of flight speed and distance; S42.

2. If , For a low risk threshold, the drone continues to perform inspection tasks while closely monitoring status changes; like , For the medium risk threshold, the drone adjusts the flight parameters to reduce the risk and tries to repair the fault, while sending a warning signal to the nest; like , initiate emergency return or alternate landing procedures; S42.

3. When the emergency return or alternate landing procedure is initiated, the UAV plans the optimal return or alternate landing path based on the current position and the location of the machine nest or the pre-equipment landing point; the path planning adopts a method based on dynamic planning, and the formula is expressed as follows: ； in, From the starting point to the sampling time position and after the The minimum cost of candidate path points, is the set of candidate path points at the previous moment, From point To point the cost; The drone flies according to the planned path and adjusts its flight attitude and speed in real time to adapt to environmental changes; the adjustment formula is as follows: ； ； in, and are the adjusted speed and acceleration, and is the expected velocity and acceleration, and is the adjustment factor; S42.

4. If the UAV is unable to return or make an emergency landing automatically, the monitoring personnel shall manually formulate an emergency response plan based on the real-time location of the UAV and the surrounding environment information provided by the remote intelligent patrol system.

7. According to claim 1, a substation drone automatic inspection method, Its characteristics are: in step S5: S51. For image data, extract basic information of the image: use a denoising method based on wavelet transform to process the image data, and use an image enhancement algorithm based on Retinex theory to process the image data; S52. Object detection based on deep learning: Use the trained convolutional neural network model to detect objects on the enhanced image; input the image into the CNN model, and the model outputs the category probability and location information of each component of the device; Equipment edge and contour extraction: For the detected equipment component area, an edge detection algorithm based on the Canny operator is used; the image gradient amplitude is calculated and the gradient direction ,in and For the image and The gradient of the direction can be obtained by convolving the image with the Sobel operator; . Then the edge pixel set is obtained by non-maximum suppression and double threshold processing. ; Then, use the morphological closing operation to connect the edge discontinuities to obtain the device contour pixel set ;Edge and contour information helps to further analyze the shape integrity and potential deformation of the device; Condition monitoring based on time series analysis: For the same equipment component in multiple consecutive frames of images, a condition monitoring model based on time series is established; The feature value in the frame image is ; Calculate the autocorrelation function of the time series and partial autocorrelation function , calculated by the Levinson-Durbin recursive algorithm; according to the characteristics of the autocorrelation function and the partial autocorrelation function, determine whether the component state is stable; if the autocorrelation function decays rapidly and the partial autocorrelation function approaches zero after a certain lag order, the component state is stable; otherwise, there may be abnormal changes; this analysis can detect dynamic changes in the state of equipment components and discover potential fault hazards in advance; S52, analyzing the infrared thermal image to extract suspected heating areas; firstly calculating the temperature histogram of the image, and determining the temperature threshold according to the distribution characteristics of the histogram ; Mark the area with temperature higher than the threshold as a suspected fever area set ; Then, the shape analysis of the suspected fever area is performed to calculate its area, perimeter, and circularity shape characteristics; The area of ​​the region is , the perimeter is , then the circularity ; By comparing the shape characteristics of the heating area with the normal equipment, it is determined whether the heating area is abnormal; this method can effectively detect the abnormal heating of the equipment and discover the hidden dangers of overheating failure in time; S53, Smoke and foreign body detection: For visible light images, use the smoke detection method based on color space analysis; convert the image from RGB color space to HSV color space, count the distribution of pixel values ​​in the HSV channel; set the smoke concentration threshold , the presence of smoke is determined based on the ratio of pixel values ​​within a specific range; if the ratio exceeds a threshold, smoke is considered to be detected and there may be a fire hazard; At the same time, a foreign body detection algorithm based on template matching is adopted; a template library of common foreign body shapes is established; for each area in the image, its similarity with the template in the template library is calculated; the template is , the image area is , similarity ; By setting the similarity threshold , screen out possible foreign body areas; this step helps to promptly detect abnormal objects in the substation environment and prevent foreign bodies from affecting equipment operation; S54, integrating the above-mentioned various detection and analysis results; marking the detected equipment defects, potential faults and abnormal conditions on the image, and the marking content includes the defect type, location and severity; Generate a preliminary analysis report, which includes inspection time, inspection equipment, a list of problems found, and recommended treatment measures; The reports are stored in a structured format, making them easier for operations and maintenance personnel to view and analyze.

8. According to the method for automatic inspection of substation by drone of claim 1, Its characteristics are: in step S6: S61. Calculate the correlation between the problem in each analysis result and the equipment in the equipment ledger; and equipment , correlation The calculation adopts the method based on text similarity and device type matching; suppose the problem description and device information are all represented as text vectors and converted through the bag-of-words model or TF-IDF method, then the text similarity ; At the same time, the type matching degree is obtained according to whether the device type matches The correlation formula is ,in and is the weight coefficient, which can be determined based on experience or data training; S62, calculate the similarity weight between the problem and the historical inspection data; for the problem and history , first extract the feature values ​​related to the problem in the historical records ; Then, calculate the difference between the problem features and the historical features ; Similarity weight ,in is the attenuation coefficient, which controls the impact of the difference on the weight; This weight reflects the degree of similarity between the current problem and the historical situation, and helps to assess the severity of the problem with reference to historical experience; S63. Score the severity of each problem based on the relevance and similarity weights; question Severity score ; Then calculate the average severity score for all issues and standard deviation ; Through severity scores and their statistical characteristics, the impact of problems on equipment operation can be quantified and the priorities of different problems can be distinguished; S64. For each device, establish a failure probability prediction model using historical inspection data; Assume that the device has been Inspections were performed at time points, and characteristic values ​​related to the equipment operation status were recorded; let the current time be , equipment in the future Probability of failure at any moment Hybrid model forecasting based on exponential smoothing and logistic regression was used; First, perform exponential smoothing on the historical eigenvalues ​​to obtain the smoothed eigenvalues ,in is the smoothing coefficient, is the original eigenvalue; Then, the smoothed eigenvalues ​​are used as input to calculate the failure probability through the logistic regression model, that is, ,in and are the parameters of the logistic regression model, which can be obtained through training with historical data; For equipment related to the current problem, the possibility of future failure is evaluated based on its characteristic values ​​and prediction models, providing a basis for preventive maintenance; S65. Generate an inspection report, which includes the following contents: Inspection time: record the time when the drone completes the inspection task; Inspection route: describes the flight path of the drone within the substation, including the equipment areas passed and the order of inspection points; Problems found and solutions: For each issue, a detailed description of the issue, severity score, comparison with historical data, and predicted probability of failure are listed; Provide corresponding handling suggestions based on the severity of the problem and the probability of failure; The inspection report adopts a structured document format, PDF or HTML, which is easy to read and archive; in the report, charts and images are used to assist explanation, making the report more intuitive and clear.

Images