Flashover Detection and Anomaly Prediction in Aerial Images of Insulator Strings in Complex Environments

The normal operation of insulator strings affects the safety and stability of the power system, and insulator string flashover is one of the important faults. In this paper, considering the flashover characteristics of insulator strings in complex environments, noise is added to the collected insulator strings to simulate the actual complex environment, and then the collected data are rotationally transformed and filtered using an improved non-local mean filtering algorithm. To accurately locate insulator strings in complex environments, the geodesic active contour correction algorithm is employed to locate and segment the image. This algorithm is developed based on the level set model, replaces the image target boundary finding problem with the problem of solving a weighted arc length minimum, and introduces a penalty term into the level set energy generalization that constrains the distance function of the level set, which effectively solves the edge leakage problem. For the segmented image of insulator strings, grayscale features and wavelet features are extracted, and then the optimized random forest algorithm is utilized for prediction analysis, and the diversity of weak learners in the random forest algorithm is increased not only through sample perturbation but also through attribute perturbation, leading to excellent generalization performance. Additionally, by combining the misclassification matrix and the a priori probability optimization algorithm, the random forest algorithm is optimized to achieve a better classification effect. Finally, the simulation of flashover discharge phenomenon characteristics and simulation experiments on a large number of insulator string images demonstrate that the proposed method has high practicality.