Refined Wind-Solar Scene Generation Research based on improved Deep Convolutional CWGAN-GP Model

A CWGAN-GP model with improved deep convolution was proposed to generate wind-solar power scenes. The deep learning algorithm is used to mine and learn the characteristics of wind and solar output data, and output of each month is accurately depicted by adding the month label information, so as to effectively generate a more accurate and practical wind-solar power output scenario set, which provides strong support for the stable operation and optimal scheduling of power system. In order to better understand and utilize the generated wind-solar power scenarios, L-ISODATA clustering method is adopted in this chapter to reasonably determine the number of initial clustering scenarios and perform cluster analysis on the generated scenarios. Typical scenes represent the main characteristics of scenery and power output, and provide a valuable reference for the subsequent scheduling optimization.