Sales Forecasting Method for Inventory Replenishment Systems of Vehicle Energy Stations Without Stockouts
针对不允许缺货的车辆能源站，提出一种结合模糊识别销售模式与深度学习预测的新方法，通过长短期记忆网络模型提高预测精度、降低成本并确保无缺货。
Accurate sales forecasting facilitates efficient and proactive replenishment, which is essential to inventory replenishment systems of vehicle energy stations that do not allow stockouts. Forecasting the sales in these systems becomes tricky because sales are highly uncertain and have different patterns at different periods. Traditional forecasting methods used in this case can greatly underestimate larger sales or overestimate smaller ones. We thus propose a novel sales forecasting method considering the fuzzy recognition technique of sales patterns and combining it with the forecasting technique based on deep learning. In this article, we first construct an index to describe the time dependence of sales. Subsequently, we determine sales patterns based on the index and fuzzy recognition. Finally, we select the long short-term memory based point forecasting model or quantile long and short-term memory based quantile forecasting model based on sales patterns, which can also fully capture the sales uncertainty and achieve the goal of reducing costs and ensuring no stockouts. Numerical experiments show that the proposed method has high prediction accuracy and can effectively reduce costs while ensuring no stockouts.