Multi-modal multi-step wind power forecasting based on stacking deep learning model

Abstract

Wind power is becoming a clean and effective energy source for electric power generation. However, the abnormity, multi-modal, and uncertainty represented in wind power data are commonly undesired. Thus, accurate wind power forecasting is a significant method for keeping the power system operations steady. To solve these issues, a multi-modal multi-step wind power forecasting model is presented. To obtain this, the density-based spatial clustering of applications with noise (DBSCAN) is improved by the k-dimensional tree (kd-tree) for detecting abnormal data. Then, the low-rank matrix fusion method fuses the wind speed, wind direction, and air density modalities for obtaining a unified representation. To further increase model accuracy, we propose a stacking deep learning model (SDLM) for overcoming the uncertainty phenomenon, which contains the bidirectional gated recurrent unit (BGRU) and leaky echo state network (LESN). The final forecasting results are acquired by a meta-learning operator. To validate the accuracy and stability of the presented approach, the inland and offshore wind farm datasets are used for forecasting. The contrastive results demonstrate that the presented model outperforms satisfactory performance in multi-step wind power prediction.