The absence of accurate and stable prediction of wind speed remains a major obstacle to the rational planning, scheduling, and maintenance of wind power generation. Currently, an extensive body of methods that aim to enhance the accuracy of wind speed prediction have been proposed. However, the majority of previous studies have tended to emphasize the structural improvement of individual forecasting models without considering the validity of data preprocessing. This can result in poor forecasting accuracy due to their failure to fully capture the effective information of the wind speed data. A new approach is proposed in this paper that successfully combines a data preprocessing technique with a linear combination method. Further, a new neural network framework is employed to determine the required combination weights to ensure improved prediction performance, thereby overcoming the drawback of the low accuracy of individual prediction models. Six wind speed datasets from Penglai are regarded as expository cases to analyze the forecasting validity and stability of the developed model. It can be concluded from the experiments that the combined forecasting system outperforms the individual models and the traditional linear combination models with higher accuracy and stronger stability.