Ultra-short-term Offshore Wind Power ProbabilityInterval Prediction and Aggregation Toward FFT Method

This paper proposes an ultra-short-term offshore wind power probability interval prediction and aggregation toward FFT method. The raw power signals are decomposed into high-frequency, medium-frequency, low-frequency, and residual components using empirical mode decomposition (EMD). A convolutional long short-term memory (CNN-LSTM) network is applied for deterministic forecasting of each component, with kernel density estimation (KDE) quantifying prediction error distributions. An innovative fast fourier transform (FFT) based method is proposed to aggregate multi-turbine probability density functions in the frequency domain, followed by inverse transformation to reconstruct time-domain cumulative distribution functions. A confidence-level analytical framework validates prediction interval reliability at $\mathbf{6 0 \%}$ and $\mathbf{9 0 \%}$ confidence levels. Simulation results demonstrate superior performance of the proposed method compared to conventional method in both interval coverage and compactness.