A combined framework based on data preprocessing, neural networks and multi-tracker optimizer for wind speed prediction

Abstract Recently, an increasing number of studies have proposed various methods for predicting wind speed to overcome the difficulties caused by the irregularity and randomness of raw data in exploring renewable wind power generation. The lack of both effective data preprocessing techniques and combined forecasting strategies has hindered the development of effective and reliable forecasting systems. In this study, a novel combined forecasting framework that simultaneously considers data preprocessing, combined forecasting, and comprehensive evaluation is presented to address the drawbacks of existing methods. To eliminate noise from raw data, complete ensemble empirical mode decomposition with adaptive noise is employed to reconstruct more reliable wind speed series for forecasting. Then, a combined forecasting module, which includes three neural networks and employs a weighted combination strategy, is designed for improving the forecasting performance, and the capability of this proposed system is verified via an evaluation module. Empirical results have demonstrated that the proposed framework not only achieves both high accuracy and stability but also provides technical support for wind power system dispatch.