A Mode-Measurement-Cooperated Interaction Framework Based on Federated Split Learning for Regional Wind Power Forecasting

With the increasing penetration of wind power generation into power grids, regional wind power forecasting errors tend to accumulate due to the clustering of wind farms, potentially affecting power dispatch. This concern can be eliminated by effectively cooperating the processes of large-scale mode simulation and data assimilation of real measurements. To address this concern, a mode-measurement-cooperation framework is proposed. In this, an adaptive Fourier operator is developed to extract partial differential features from mode data, followed by a dense-sparse convolution to fuse simulated modes with measurements from multiple wind farms. Last, to accommodate the interaction between centralized server and multiple decentralized wind farms, federated split learning is adopted as the core interaction mechanism, which jointly ensured the information interaction and privacy preservation. The proposed framework is validated based on practical data from 156 wind farms in a provincial region. Compared with other state-of-the-art methods, the mode-measurement-cooperated mechanism is observed to increase the structure similarity index measure of wind vector image by 4.348% and reduce the normalized mean absolute error of regional wind power forecasting by 4.071%. In addition, the advantages of the proposed framework on the privacy preservation and mutual computational friendship also confirmed the practical significance of the proposed framework.