Deterministic and probabilistic wind speed forecasting using decomposition methods: Accuracy and uncertainty

Wind energy is gaining in importance as a source of renewable energy. However, because wind speed is intermittent, integrating wind energy into the grid requires accurate forecasts. This paper tests several forecasting models based on decomposition, using data from two representative sea-land sites, Fujian and Inner Mongolia. It also analyzes the diverse uncertainty perspective of probability distributions and the role of causal factors. The main finding is that a method based on extreme machine learning achieves greater forecast accuracy than other decomposition-based methods. A further finding is that the uncertainty associated with forecasts can vary as a function of the model, and more specifically, the decomposition method. The results of experiments show that the optimal forecasting model for Fujian and Inner Mongolia wind speed is EMD-SE-WT-PSO-ELM and CEEMDAN-SE-WT-PSO-ELM respectively, with error indicator improvement rates both exceeding 27.99 %. However, in the long-term forecasts for the next 4 h, VMD-SE-WT-PSO-ELM gave the best forecasts, with goodness-of-fit above 90 % in all cases.