A Photovoltaic Power Scenario Generation Method Based on HiLo-WGAN

To address the challenges posed by the uncertainty of photovoltaic output on power system operation and scheduling, this paper proposes an improved photovoltaic scenario generation model using a generative adversarial network with gradient penalty. First, the Wasserstein distance is used as the loss function to construct the generator and discriminator models, incorporating a gradient penalty in the adversarial loss calculation. Second, a residual network's identity mapping structure is employed to adaptively adjust the architectures of the generator and discriminator, with the HiLo attention mechanism (high frequency and low frequency, HiLo Attention) integrated into the discriminator. This mechanism extracts high and low-frequency features from the input samples and combines these features to form the final attention output. Finally, through the optimized adversarial training process of the GAN, accurate prediction of photovoltaic output is achieved. The proposed method is validated using a real photovoltaic power dataset, and the results show that it has excellent scenario generalization capabilities and significantly improves the accuracy of photovoltaic power prediction.