Over the past two decades, the grey model (GM) and its extensions have shown to be an effective tool to deal with short-term time series (STTS) data. To further improve the accuracy of GM models, this paper proposes a novel GM modelling procedure based on a fuzzy-set concept. In the procedure, STTS data is fuzzified by measuring the forces between existing data and a newly coming datum to form fuzzy time series for building GM models, and final predictions are defuzzified by aggregating the predictions of the GM models with the proposed weights. The experimental results of a real case indicate that the proposed procedure can further improve the accuracy of GM models in general.