Electronics, Год журнала: 2024, Номер 13(23), С. 4829 - 4829

Опубликована: Дек. 6, 2024

Wind speed, wind direction, humidity, temperature, altitude, and other factors affect power generation, the uncertainty instability of above bring challenges to regulation control which requires flexible management scheduling strategies. Therefore, it is crucial improve accuracy ultra-short-term prediction. To solve this problem, paper proposes an prediction method with MIVNDN. Firstly, Spearman’s Kendall’s correlation coefficients are integrated select appropriate features. Secondly, multi-strategy dung beetle optimization algorithm (MSDBO) used optimize parameter combinations in improved complete ensemble empirical mode decomposition adaptive noise (ICEEMDAN) method, optimized decompose historical sequence obtain a series intrinsic modal function (IMF) components different frequency ranges. Then, high-frequency band IMF low-frequency reconstructed using t-mean test sample entropy, component decomposed quadratically variational (VMD) new set components. Finally, Nons-Transformer model by adding dilated causal convolution its encoder, components, as well unreconstructed mid-frequency IMF, inputs results perform error analysis. The experimental show that our proposed outperforms single combined models.

Язык: Английский