PosiWGAIN-GP-L2:A Novel Method for Missing Wind Speed Data Imputation Based on Generative Adversarial Imputation Networks

Wind speed data is crucial for energy planning, renewable energy utilization, and power system optimization. However, missing and uncertain data frequently occur, making efficient data imputation a key challenge. This paper proposes a novel imputation method, positional encoding wasserstein generative adversarial imputation networks with gradient penalty and L2 regularization (PosiWGAIN-GP-L2), to address these issues. The method integrates Wasserstein distance into the GAIN model, adds gradient penalties to satisfy the Lipschitz condition, and uses positional encoding to capture temporal wind speed features. To address training difficulties caused by high data dimensionality, an L2 regularization term is introduced to constrain the generator. Experimental results on wind speed data with varying missing rates across four seasons show that the proposed method significantly improves imputation accuracy, adapting well to different seasonal patterns and missing rate changes.