Adaptive Gradient Penalty-Based Wasserstein GAN for Photovoltaic Power Prediction

To address the challenges of photovoltaic (PV) power forecasting and low utilization caused by volatility and intermittency, this study proposes an improved WGAN-GP method based on adaptive gradient penalty for fast and accurate PV power prediction. The traditional WGAN-GP model suffers from slow convergence and high sensitivity to hyperparameters, limiting its practicality. The improved method dynamically adjusts the gradient penalty coefficient by monitoring the discriminator gradient deviation, significantly improving convergence speed while maintaining comparable sample quality to traditional methods.