Wind speed forecasting is fundamental to the dispatching, controllability, and stability of the power grid. As a challenging but essential work, wind speed forecasting has attracted significant attention from researchers and managers. However, traditional forecasting models sometimes fail to capture data features owing to the randomness and intermittency of wind speed, and the models always focus only on improving accuracy, which is one-sided. Motivated by these problems, in this study, a novel hybrid forecasting system consisting of three modules (a data preprocessing module, optimization module, and forecasting module) is developed to improve the forecasting accuracy and stability. In the data preprocessing module, an effective denoising technique is used to produce a smoother sequence. A fuzzy time series method optimized by a multi-objective differential evolution algorithm that balances the conflict between forecasting accuracy and stability is developed in the next two modules to perform the forecasting. Several experimental results covering different models and wind-speed time intervals all indicate that our proposed hybrid forecasting system can achieve both satisfactory accuracy and stability with a mean absolute percentage error of below 4% in wind speed forecasting. Moreover, the statistical hypothesis test, forecasting effectiveness and grey relational degree are discussed which also demonstrate the great performance of proposed system.