A multi-view and multi-scale transfer learning based wind farm equivalent method. (May 2020)

• Record Type: Journal Article
• Authors: Han, Ji; Miao, Shihong; Li, Yaowang; Yang, Weichen; Zheng, Tingting
• Is Part Of: International journal of electrical power & energy systems. Volume 117(2020)
• Publication Date: 2020-05
• Subjects: Wind farm equivalent -- Transfer learning -- Multi-view and multi-scale fuzzy C-means (VS-FCM) clustering -- Refined composite multi-scale entropy (RCMSE) -- Q-learning; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2019.105740
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: Refined composite multi-scale entropy (RCMSE) method is used to extract feature of WT, and multi-scale concept is used in multi-view FCM (V-FCM) to form a new clustering algorithm: multi-scale V-FCM (SV-FCM). Transfer learning is applied to SV-FCM, and transfer SV-FCM (TSV-FCM) clustering algorithm is proposed, which is used for clustering WTs. Transfer Q-learning algorithm is used to optimizing collector network parameters, which could obtain collector network parameters rapidly and accurately. Abstract: With the increasing capacity of wind farm (WF), detailed WF model is not appropriate for power system studies, and the equivalence of WF with the required accuracy level poses a complicated technical challenge. In this paper, a multi-view and multi-scale transfer learning based WF equivalent method is proposed. Three steps are taken in this method. (a) Extract feature from active and reactive power of wind turbine (WT) using refined composite multi-scale entropy (RCMSE), on this basis, construct clustering indicator considering view and scale two aspects. (b) Aiming at the feature of the clustering indicator, a multi-view and multi-scale fuzzy C-means (VS-FCM) clustering algorithm is proposed, and transfer learning is used in it for better WTs cluster performance. (c) transfer Q-learning is adopted to optimize the parameters of collector network for each equivalent WT, so as to improve parameter optimization efficiency. To verify the effectiveness of the proposed method, an actual system in East Inner Mongolia of China is utilized for case study. Simulation results shows that the dynamic characteristics and the robustness of the proposed model perform a good behavior in different wind scenarios and voltage sag levels, besides, the method has an advantage in the efficiency of simulation time and parameter optimization.