A new framework for the medium and long-term stochastic scenario generation of wind power output

Medium and long-term wind power output scenarios are of great significance for power system operation and planning. However, the complex spatiotemporal features among multiple wind farms and the high-dimensional characteristics of medium and long-term output data make it difficult to generate the scenarios properly. In this paper, a new medium and long-term wind power scenario generation framework is proposed, which presents a manifold learning method to extract joint output features of multiple wind farms, and combines deep learning and statistical methods to generate medium and long-term wind power output scenarios. Firstly, the uniform manifold approximation and projection (UMAP) method is adapted to reduce the dimensionality of the historical high-dimensional scenario set of wind power output to a low-dimensional manifold, while preserving the local neighborhood relationships and global spatiotemporal structures. Secondly, to obtain multiple states of daily wind power scenarios and to construct a state transition probability matrix, an affinity propagation clustering algorithm (AP) based on the Davies-Bouldin (DB) index is designed to improve clustering robustness and generalization. Finally, based on the state transition probability matrix and the set of daily scenarios generated by deep learning, a medium and long-term scenario splicing method for multi-wind farm output is proposed using Markov Chain Monte Carlo (MCMC) method. The simulation results based on the output data of 8 actual historical wind farms verify the effectiveness of the proposed method. This method has strong applicability and can be used for most wind power output conditions.