Wind Power Prediction Based on CEEMDAN-VMD Two-Level Decomposition and CNN-BiLSTM Hybrid Network

Abstract

With the increasing penetration of wind power in the electric power system, the impact of wind power prediction accuracy on the safe and stable operation of the grid has become increasingly prominent. In view of the limitations of traditional prediction methods, which struggle to effectively handle the non-stationarity and multi-scale characteristics of wind power sequences, this paper proposes a hybrid prediction model based on the combination of signal decomposition and deep learning. This method adopts a two-level signal processing mechanism consisting of fully adaptive noise ensemble empirical mode decomposition and variational mode decomposition to extract multi-scale features from the original power sequence. It combines a hybrid architecture of convolutional neural networks and bidirectional long short-term memory networks to capture local features and long-term dependencies of the sequence, respectively. Experimental verification using actual wind farm operation data shows that the proposed method can effectively improve the accuracy of wind power prediction, exhibiting good adaptability and stability under different fluctuation scenarios. It provides reliable technical support for the scheduling and operation of high-proportion wind power integration into the grid.