Research And Application Of Key Technologies For Autonomous Inspection Of Overhead Transmission And Distribution Power Line By Unmanned Aerial Vehicles

In recent years,unmanned aerial vehicles(UAVs)have been widely applied in the field of power grid equipment inspection.Multirotor drones,known for their hover-capable and small size,have been increasingly used for efficient inspection of equipment by carrying various types of sensors,significantly improving work efficiency and quality.Specifically,in the inspection of overhead transmission and distribution lines,they have gradually become the primary means for equipment operation and maintenance.In this thesis,the inspection of overhead transmission and distribution lines using multirotor drones will be referred to as UAV line inspection,and the multirotor drones used for inspection will be referred to as inspection drones.However,the current UAV line inspection technology also faces a series of technical challenges.In terms of inspection drone design,the inspection drones used in the industry exhibit varying degrees of issues,including poor portability,low endurance,and limited sensor types.Regarding autonomous flight technology,the existing techniques require extensive pre-work,such as high-precision 3D modeling and route planning validation,resulting in low overall work efficiency and vulnerability to external environmental changes.Moreover,high reliance on mobile digital networks during flight imposes demanding operating conditions.In terms of inspections of key components,the lack of targeted data collection for critical component targets leads to suboptimal quality of inspection data,thereby impacting defect detection accuracy.In terms of inspection data processing,the overall accuracy of defect identification is relatively low,with the majority of massive inspection data still heavily reliant on manual processing,predominantly in an offline centralized manner,resulting in poor real-time data analysis.Therefore,although UAV inspection technology has become an important means to enhance the efficiency and quality of transmission and distribution line inspections,there is still significant room for improvement to meet the urgent demand for high-quality development in the operation and maintenance of transmission and distribution lines.To address the aforementioned issues,this thesis focuses on the following four research areas:(1)Design of Multirotor Inspection UAV: In response to the issues of poor portability and low endurance of current UAVs used for power grid inspection,this study focuses on the design technology of compact and long-endurance inspection UAVs.It optimizes the aerodynamic performance of the UAV structure and designs an integrated fuselage structure,followed by dynamic simulation analysis.A PID flight power control method based on disturbance observer is proposed to address the issues of unstable airflow affecting the attitude and position of small UAVs,thereby enhancing the robustness of the controller.Considering the characteristics of the inspection UAV’s fuselage structure and demanding Electromagnetic Compatibility(EMC)requirements,hardware circuit design is conducted to meet the characteristics of high current,high reliability,and rapid response.To address the problem of single-sensor integration on inspection UAVs,an innovative design of an integrated payload,combining sensors such as Li DAR,visible light,and infrared imaging,is proposed.By achieving time synchronization alignment among the sensors through GNSS and pulse signal timing,the UAV’s flight control system implements data fusion from multiple sensors using the Extended Kalman Filter(EKF)architecture.This supports the collection of multiple types of inspection data within a single flight mission,thereby improving inspection efficiency.(2)Autonomous Flight Control Technology for Inspection UAVs: To address the limitations of existing autonomous flight technologies that require high-precision 3D modeling and route planning in advance,this thesis proposes a deep learning-based3 D point cloud data classification algorithm.A point cloud classifier is constructed to achieve real-time recognition of key target objects and real-time construction of the3 D scene.Additionally,an INS-GNSS deeply coupled multi-source data positioning algorithm is proposed to solve the problem of high-precision positioning in weak mobile digital network environments,enabling autonomous positioning of inspection UAVs.The study explores a wire-following technique based on multiple sensors and a method for optimal route analysis and optimization,enabling autonomous navigation of inspection UAVs and facilitating autonomous flight.Furthermore,through the fusion of 3D Li DAR and visual image technologies,the inspection UAV achieves autonomous obstacle avoidance.A method for calculating the optimal safe flight path of the inspection UAV is also proposed to further enhance the safety of autonomous flight.(3)Online Recognition of Key Components in Power Transmission and Distribution Lines: To address the issue of low-quality inspection data and its impact on defect detection accuracy due to the lack of targeted data collection for key components,this study investigates the use of an improved YOLOv5 s algorithm for online localization and recognition of critical targets such as insulators,fittings,and power lines.It proposes a hierarchical optimization model based on visual features and an attention-based algorithm for detecting occluded targets,enhancing the feature extraction and recognition capabilities of the model.Several data augmentation techniques are employed to expand the training set and reduce overfitting on the original dataset,thereby assisting in model training and improving its performance.The study explores visible light image target localization techniques to control the inspection drone’s movement to predefined positions and adjust the gimbal angle based on the recognized targets,allowing for the acquisition of more accurate and clear inspection data.(4)Online Detection of Typical Defects and Tree Hazards in Power Transmission and Distribution Lines: To address the issue of low accuracy in automatic defect recognition from inspection data,this study adopts the Focal Loss-enhanced Efficient Net algorithm to achieve highly accurate detection of typical defects in insulators and fittings.For long and slender targets like power lines,the U-Net algorithm,combined with elastic transformation augmentation methods,is employed to improve network segmentation performance and detect wire breakage and strand scattering defects.The study proposes an adaptive defect dataset feature enhancement algorithm based on image saliency detection to overcome the problem of feature masking caused by erasure-based image augmentation.A real-time measurement and calculation model for detecting tree hazards in overhead power transmission and distribution lines is developed.To address the issue of predominantly offline centralized data processing and poor real-time performance,the study investigates the online deployment of deep learning models using Tensor RT technology to enhance the network model’s operational efficiency on onboard computing units,enabling high-performance edge computing to meet the requirements of online recognition and detection.Building upon the aforementioned research,practical engineering applications and verifications were conducted in actual scenarios of typical power transmission and distribution lines in Guangdong Province,including 500 k V,220 k V,110 k V,and10 k V lines.The results of these applications demonstrate the practical applicability and effectiveness of the key technologies developed in this study.