Intelligent prediction of 110kV insulator lightning flashover criteria based on random forest

The lightning flashover criteria of insulators vary significantly under different conditions, and it is impractical to exhaustively enumerate all the criteria through extensive tests. Therefore, this paper aims to explore the intelligent prediction of lightning flashover criteria for 110 kV insulators using a data-driven approach. Initially, a test database comprising 4,978 high- and low-altitude test data of 110 kV insulator lightning impulse flashover is established. Subsequently, relevant characteristic values are extracted from the database as inputs for the prediction model. Various machine learning algorithms, such as the BPNN, SVM and RF algorithms, are employed to construct prediction models for the U50% and volt-time characteristics of 110 kV insulators under lightning impulses. The results demonstrate that the RF algorithm yields an average absolute percentage error of merely 1.17 % for the U50% prediction model. Additionally, the RF and BP algorithms achieve the highest prediction accuracies of 10.7 % and 6.5 %, respectively, for the volt-time characteristics at high- and low-altitude. This validates the feasibility of substituting many traditional tests with more accurate flashover criteria predicted through a data-driven approach. This paper provides a novel concept for predicting the lightning impulse flashover criteria of insulators under different working conditions. Reducing the need for repetitive tests is expected to acquire the high-precision intelligent prediction of insulator lightning impulse flashover criteria.