Insulator leakage current prediction based on generative adversarial networks and optimized support vector regression with crisscross optimization algorithm

Current methods for insulator leakage current prediction usually cannot guarantee satisfactory accuracy. To address this issue, a novel prediction method is proposed based on gradient penalized Wasserstein generating adversarial network (WGAN-GP) and an improved support vector regression (SVR) model. The proposed model can: 1) learn the distribution pattern of the real data and generate high-quality training data; 2) optimize parameters of SVR model through the crisscross optimization algorithm (CSO), and 3) improve the prediction accuracy. Owing to the unique gradient penalty, the WGAN-GP network is firstly used to generate high-quality training samples and achieve data augmentation. Then CSO is applied to optimize the model parameters of SVR and thus an improved prediction model is constructed. Finally, the generated data and optimized parameters are applied in the proposed method to predict the insulator leakage current. Experimental results show that the proposed method outperforms the state-of-the-art models in all evaluation indexes and improves the prediction accuracy.