Ultra-short-term Wind Power Prediction Based on Deep Ensemble Learning Model Using Multivariate Mode Decomposition and Multi-objective Optimization

To address the issue of ultra-short-term wind power prediction, a novel prediction model is proposed based on multivariate variational mode decomposition (MVMD), multi-objective crisscross optimization (MOCSO) algorithm and blending ensemble learning. In the data processing stage, to maintain synchronization correlation and ensure matching of intrinsic mode fuctions (IMFs) number and frequency, the MVMD method is used to decompose the multi-channel original data synchronously. Considering the insufficient comprehensiveness, inaccuracy, and low robustness of the single machine learning model, a blending ensemble learning model is proposed to combine multiple deep learning networks using MOCSO dynamic weighting. The prediction results of RNN, CNN and LSTM are dynamically weighted, integrated, and then optimized by MOCSO to improve the prediction accuracy and stability. Experimental results show that the proposed model is not only effective, but also significantly superior to other prediction models.