The Study of Cluster Predication Method on Sales Forecast Based on Residual Error Modified GM (1, 1)

The amount of sales un-house is important basis for inventory management of commerce enterprises. Through carefully analyzing about previous researching results, we find that tradition forecasting methods, such as time series analysis, regression analysis, Kalman filtering and the predictions of neural networks, have some defects in large information demanding, the numerical instability and insensibility to environment changes. So, based on GM (1,1) model and combining calamity grey prediction at residual hour, this paper establishes a REM-GM (1,1) model and with the aid of cluster prediction method successfully forecasts the amount of sales un-house of commerce enterprises. The empirical studies observe that the model of un-house forecasting, no matter whether one step forecast or multi-step long time forecast, has a more remarkable prediction precision.