将季内振荡与数值天气预报结合用于15天风电功率预测

Extending the timescale of wind power forecasting (WPF) is vital for grid management and market operations in renewable-dominated power systems. However, the substantial dependence of WPF on numerical weather prediction (NWP) presents a considerable challenge. The iterative operations of NWP based on short-term data amplify its inherent uncertainty, reducing its accuracy beyond 10 days. To address this, intra-seasonal oscillation (ISO) is introduced to capture longer-term and larger-scale meteorological patterns, leading to the proposition of an ISO-NWP integrated framework for 15-day WPF. Firstly, a historical spatiotemporal localization model for teleconnections (TC) is developed, which connects distant weather changes and wind power fluctuations under ISO. Subsequently, a TC automatic selection network is designed as the encoder of the WPF network, which integrates ISO-NWP and computes dynamic weights of TC through tensor inner products. Following this, a trend-detail sequential network is designed as the decoder, enhancing the ability to fit long wind power sequences by learning both trends and detailed fluctuations. Lastly, the effectiveness is validated using real data from 3 wind farm clusters, encompassing 26 wind farms across 3 provinces in China.