Multi-objective wind power scenario forecasting based on PG-GAN

Accurate scenario forecasting of wind power is crucial to the day-ahead scheduling of power systems with large-scale renewable generation. However, the intermittence and fluctuation of wind energy bring great challenges to the improvement of prediction accuracy. Aiming at precisely modeling the uncertainty in wind power, a novel scenario forecasting method is proposed in this paper. First, Progressive Growing of Generative Adversarial Networks is leveraged to capture the complex temporal dynamics and pattern correlations. Second, wind power scenarios of the forecast day are achieved by solving a multi-objective scenario forecasting problem with progressive optimization-based Non-dominated Sorting Genetic Algorithm III. Finally, a real wind power dataset and a real power system scheduling problem are applied to justify the effectiveness of the research. Experimental results based on the dataset indicate that our method produces high-quality scenarios with richer details compared with existing research even if the given point forecast is inaccurate. Besides, different amounts of scenarios can be provided without sacrificing time efficiency, which follow the actual trend of wind power consistently and demonstrate great superiority in three evaluation metrics. Moreover, experimental results of the scheduling problem also prove that our method outperforms the others on expected total costs and unmet load amounts.