Abstract: Solar energy exhibits randomness, volatility, and intermittency, making it difficult to directly integrate photovoltaic power generation into the grid. Therefore, research on PV power prediction is essential to enhance the absorption capacity of PV generation and the reliability of the grid. This paper proposes a short-term PV power prediction method based on an improved generative adversarial network and a gated recurrent unit. First, the original dataset is categorized into three weather types-sunny, cloudy, and rainy-using the K-means clustering algorithm (K-means) for similar-day classification. Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) is employed to perform data augmentation for cloudy and rainy days, which have limited data availability. Next, the whale optimization algorithm (WOA) is used to optimize the gated recurrent unit with attention mechanisms (GRU-Attention) for PV power prediction. The algorithm fine-tunes the initial learning rate, batch size, and number of hidden layer nodes to improve prediction accuracy. To validate the effectiveness of the proposed method, a power prediction algorithm model was developed in a Python environment. Comparative experiments were conducted with the non-optimized GRU-Attention network and a convolutional long short-term memory neural network with attention mechanism (CNN-LSTM-Attention). The results demonstrate that the proposed method achieves maximum reductions of 6.47% and 5.18% in normalized root mean square error (NRMSE) and normalized mean absolute error (NMAE), respectively, across all three weather conditions. After data augmentation, the prediction errors for rainy and cloudy days have further decreased.