Deterministic wind energy forecasting: A review of intelligent predictors and auxiliary methods.

• A comprehensive review of intelligent predictors and two auxiliary methods in hybrid forecasting models. • Discussion on merits and limitations of intelligent predictors for wind energy forecasting applications. • Sources of diversity and ensemble strategies in ensemble learning are summarized and clarified. • Overview of underlying theories of metaheuristic optimization algorithms and their implementation details. Recent developments in renewable energy have highlighted the need for rational use of wind energy. Accurate prediction of wind speed and wind power is recognized as an essential part in realizing energy balance and scheduling decisions of power generation. In recent years, various wind energy forecasting models have been successfully proposed. Among them, intelligent models occupy an irreplaceable dominance and have tremendous potential due to their accuracy and robustness. This paper gives a broad literature survey of the intelligent predictors in the field of wind energy forecasting, including four types of shallow predictors (artificial neural network, extreme learning machine, support vector machine, and fuzzy logic model) and four types of deep learning-based predictors (autoencoder, restricted Boltzmann machine, convolutional neural network, and recurrent neural network). Their theoretical backgrounds, applications, merits, and limitations are thoroughly discussed. Then, two commonly used auxiliary methods for hybrid intelligent models are reviewed, i.e., ensemble learning and metaheuristic optimization. The ensemble learning models are categorized by the sources of diversity and ensemble strategies. According to the specific optimized objects, the metaheuristic optimization algorithms are classified into two groups. Moreover, the general process of metaheuristic optimization and differences between single-objective and multi-objective algorithms are also clarified. A group of representative models is summarized to show the frameworks of mainstream predictive models in artificial intelligence. Finally, this paper gives three possible development directions of wind energy forecasting for subsequent research. [ABSTRACT FROM AUTHOR]