Abstract: Accurate short-term demand forecasting is crucial for the production plan development, but a shorter forecasting period implies that the product demands are more unstable and, thus, that ascertaining their developing trends is difficult. Therefore, using large historical observations to build forecasting models may not result in favorable forecasting performance. Prediction methods based on the latest limited data have thus become vital for maintaining management efficiency and competitive advantage. Grey system theory is a technique for resolving this difficulty. However, the conventional grey model is designed for time-series data featuring equigaps, which limits its application scope. Although in the current published research, scholars have proposed some non-equigap grey models; however, the newest datum is usually weakened to alleviate the randomness of data in these models, which may result in a larger prediction error. To conquer these shortcomings, this study introduces linear extrapolation to modeling procedures for emphasizing the importance of the newest datum and then proposes an improved non-equigap grey model. In the experiments of two real cases, the proposed method exhibits favorable forecasting performance, indicating a feasible solution for small non-equigap data forecasting.  

Key words:  , Grey System Theory, Small Data Set, Forecasting, Short-term Demand, Non-equigap