Wind Power Ultra-short-term Interval Prediction Based on an Adaptive Kernel Density Estimation Algorithm

Aiming at the issue of insufficient local adaptability in kernel density estimation algorithm for wind power interval prediction, we propose an adaptive kernel density estimation algorithm for wind power ultra-short-term interval prediction. Firstly, a long short term-memory wind power ultra-short-term prediction model is constructed based on the improved sequence to sequence (Seq2Seq) architecture with a dual-stage attention mechanism for wind power deterministic prediction. Subsequently, an proposed adaptive kernel density estimation algorithm is utilized to estimate the kernel density of the historical wind power prediction error. The proposed algorithm is capable of adaptively adjusting the bandwidth according to the kernel density estimation of the prediction error. Finally, the cumulative distribution function of the prediction error is obtained by integrating the kernel density estimation results. In addition, the upper and lower quantiles under different confidence coefficients are calculated, and the ultra-short-term prediction intervals of wind power under the corresponding confidence coefficients are obtained by adding the quantiles and the deterministic prediction values. Simulation analysis based on the data from a wind farm in Northwest China demonstrate that the proposed adaptive kernel density estimation algorithm exhibits superior adaptive ability to wind power prediction error data. Moreover, the effectiveness of this wind power interval prediction model is validated through these experiments.