After analyzing the deficiency of traditional exponential smoothing prediction model, a novel prediction model with the concepts of dynamic smoothing parameter and initial value was proposed and applied to prediction compensation-control in small-batch production. In the presented model, with an optimal algorithm based on steepest-descent, the dynamic smoothing parameter and initial value were adjusted with new sample of the time-series automatically. Moreover, the presented model was used as the key technology of quality predicting in machining plungers of oil pump mouth, the practical results showed that compared with auto-regressive(AR), gray and traditional exponential smoothing model, the presented one is higher in precision and better in adaptation.