An On-Line Neuro-Forecasting Method with Corrective Model in Ajou Neuro- Advanced Planning Scheduling Project

摘要

The AN-APS (Ajou Neuro-Advanced Planning & Scheduling) Project is aimed at developing an intelligent forecasting, inventory replenishment planning, distribution resource planning, and integrated production planning system. In this presentation, we focus on developing an efficient forecasting method with corrective model. The situation considered in this study requires on-line model identification of a huge number of forecasting models. A new multi-stage forecasting method is presented and integrated it with an adaptive distribution resource planning. The proposed multi-stage forecasting method uses neural network approach and multi-tuning procedure simultaneously. The multistage tuning procedure, considered as a corrective prediction model, is based on step-wise multiple regression or additional neural network model depending on data. Statistical testing procedure is employed for identification of proper corrective model. The multi-stage tuning procedure is designed for improving forecast accuracy by using additional corrective model. Based on the forecasting results, an adaptive inventory replenishment planning is designed. The proposed system applied for a real food distribution system and could be evaluated as an effective methodology for the given application domain