A transfer learning-based scenario generation method for stochastic optimal scheduling of microgrid with newly-built wind farm. (February 2022)

• Record Type: Journal Article
• Authors: Hu, Jinxing; Li, Hongru
• Is Part Of: Renewable energy. Volume 185(2022)
• Page Start: 1139
• Page End: 1151
• Publication Date: 2022-02
• Subjects: Scenario generation -- Transfer learning -- Wind data -- Joint distribution adaption -- Wind farms -- Stochastic optimal scheduling; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2021.12.110
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: In the stochastic optimal scheduling of microgrid with multiple wind farms, the accurate description of uncertainties is a critical issue. Scenario generation provides an effective way to represent the strong randomness and interdependence between wind speeds. However, there may be very limited data or no historical information in the beginning stage of a newly-built wind farm operation, which will lead to the inaccuracy of scenario generation and thus affect the reliability of decision results. In this paper, considering that multiple wind farms in the adjacent areas may have similar weather conditions, a novel transfer learning-based scenario generation method is proposed to utilize the historical information from other existing data-rich farms for generating wind speed scenarios of the new farm. The scenario generation tasks are constructed as a cross-domain adaption problem. To model the target wind speed, joint distribution adaption (JDA) is adopted to explore the underlying relationship between multiple source farms and the target farm. Experimental results show that the scenarios generated by our proposed method can better describe the properties of target wind speed, and the microgrid scheduling results can be more reliable in the case of very limited data. Highlights: A novel scenario generation method based on transfer learning is developed. Joint distribution adaption is adopted to model wind speed in the new wind farm. Spatially correlated scenarios with high accuracy are generated effectively. Microgrid scheduling results are more reliable in the case of very limited data.