Short-run Wind Power Combination Projection Model based on CEEMDAN-TPA-TCN-MRFO

Abstract

Wind power is playing an increasingly critical role in new energy sources. In order to modify the precision of short-term wind power prediction, this thesis proposes a combined model for short-term wind power prognosticate based on Complete Ensemble Empirical Mode Decomposition of Adaptive Noise (CEEMDAN), Time Pattern Attention mechanism (TPA), Time Convolution Network (TCN) and Multi-strategy Manta Ray Foraging Optimization (MRFO). Firstly, this thesis uses CEEMDAN algorithm decompose the wind power sequence, and obtain the sub-sequence components. Combined with key meteorological variable data, this thesis constructs a training set. Then, it optimizes the convolutional network hyperparameters with MRFO. Finally, the final forecasting value is obtained after reconstructing the forecasting results by predicting sub-sequence components respectively with TPA and TCN. The practical cases reveal that the propound combined forecasting model can better predict the wind power change trend. Compared with other methods, this model can significantly improve the short-term presaging accuracy of wind power.