Novel wind speed forecasting model based on a deep learning combined strategy in urban energy systems

Abstract

Effective wind speed forecasting has great significance for urban energy system operations and the construction of low-carbon cities. However, most previous research has focused only on data in the same frequency, limiting forecast performance to some extent. In this study, considering the value of mix-frequency data, a deep learning combined model based on mixed frequency modeling is developed to improve forecast effectiveness. Firstly, a data preprocessing module is designed to decompose and reconstruct the original low- and high-frequency wind speeds. Then, a mixed frequency modeling module, comprising four mixed data sampling models and four machine learning models, is proposed to achieve mixed frequency wind speed forecasting. Further, the optimal sub-models are determined based on a newly developed evaluation index. Finally, a deep-learning-based non-linear combination forecasting module is developed to realize wind speed forecasting by taking full advantage of optimal sub-models to increase forecasting performance and guarantee the developed model’s accuracy and stability. Furthermore, a scientific and comprehensive evaluation module is established. Four experiments and eight discussions based on real wind farms demonstrate that the developed model can significantly enhance wind speed forecasting performance, accelerate the construction of low-carbon cities, and improve the sustainable and resilient development of urban energy systems.