A new arXiv preprint tackles voltage time-series forecasting when anomalies are present. The paper points out that recursive forecasting on high-dimensional data increases the complexity of prediction, and that long-term dynamics become harder to model as a result.
To address this, the study proposes a Wasserstein GAN (WGAN) model with Lipschitz control and a progressive kernel critic strategy. The abstract does not provide further details on the architecture, training procedure, or experimental outcomes.
Because only the opening of the abstract is available, the article cannot assess the model's performance or compare it with other approaches. The contribution is presented as a method for handling the challenge of long-term dynamics in high-dimensional forecasting under anomalies.