Experimental Database and Prediction Model of Insulators Flashover Under Lightning Impulse

Lightning flashover criteria of insulators and air gaps are the basis of the external insulation design for transmission lines. However, the mechanisms of long air gap discharges remain unclear. This paper aims to establish a comprehensive database of lightning impulse tests for insulators/air gaps based on extensive experiments conducted over the past twelve years, making it feasible to predict the flashover characteristics of insulators/air gaps accurately through machine learning algorithms. Firstly, a comparative analysis of experimental results from various institutions over the years reveals that the flashover characteristics are significantly influenced by many factors, including gap/insulator length, voltage waveform, insulator materials, and meteorological conditions. Subsequently, a publicly accessible insulator/air gap flashover test database is constructed to realize the real-time storage, management, and analysis of test data. Ultimately, the input features are selected by correlation analysis. The intelligent prediction models for the lightning flashover characteristics of 220kV insulators are developed by algorithms, including SVR (Support Vector Regression), ELM (Extreme Learning Machine), XGBoost (eXtreme Gradient Boosting), and RF (Random Forest). The results demonstrate that the SVR algorithm achieved a mean absolute percentage error (MAPE) of merely 2.19% in predicting U50% compared to the experimental results. Meanwhile, XGBoost achieved the lowest error of 18% in volt-time characteristics. It shows promising potential for more accurate criteria through the use of the test database and advanced algorithms.