Wind speed forecasting is an important task in large-scale wind power integration that can eliminate the harmful influence caused by its inherent intermittence and volatility. To achieve high-precision wind speed forecasting, hybrid systems that combine artificial intelligence algorithms have been widely employed. However, in most previous studies, in the learning process of wind speed forecasting using hybrid systems, two open challenges arise: it is hard to ensure that the main features of wind speed data can be completely extracted using simple data preprocessing, and attempting to enhance only the accuracy while ignoring the stability is insufficient in practical applications. In this study, a novel hybrid forecasting system is successfully proposed to solve the abovementioned issues, with the following novel contributions: (i) a new data preprocessing algorithm is developed based on the proposed hybrid data preprocessing strategy, which combines the advantages of each meritorious component and appears to be a promising data preprocessing method; and (ii) the Elman neural network model, improved by our newly proposed multi-objective satin bowerbird optimizer algorithm, is successfully developed, which provides a great contribution to the excellent forecasting performance in terms of accuracy and stability. To verify the forecasting effectiveness of the system, several forecasting cases based on eight wind speed datasets are presented in this study, and the results reveal that the proposed system has better forecasting accuracy and stability than other benchmark models and can be used to enhance the utilization efficiency of wind energy as well as other fields.