As an important clean energy source, wind power has strong influence on the safe operation of the power grid. Because its output depends directly on the wind speed, it is of great significance to accurately forecast the short-term wind speed. However, many previous studies have two shortcomings: they only focus on the original data without extracting the information of the error sequence; most of them are based on single optimization and neglected the importance of increasing the accuracy and stability simultaneously. Therefore, a novel hybrid model based on data feature extraction and multi-objective optimization combining the original wind speed series with the error sequence non-linearly is proposed for short-term forecasting. It consists of two steps: (a) obtain error sequence by the preliminary prediction, and (b) Combining the error sequence with the preliminary results to obtain the final predicted value. In addition, a hybrid Elman neural network (ENN) optimized by multi-objective grey wolf optimization (MOGWO) is proposed to achieve high accuracy and stability at the same time. To analyze the performance, four datasets collected from different wind fields are utilized as an example. The results demonstrate the proposed forecasting model reaches superior accuracy and stability when compared with other contrast models. (C) 2019 Elsevier Ltd. All rights reserved.