Nonparametric Average Inventory Management Based On Censored Demand

In this paper,we study an inventory management problem when the demand distribution is unknown,and propose an algorithm named nonparametric average inventory management(NAIM)based on sales data.For this inventory management problem,Huh(2009)proposed an online learning algorithm named adaptive inventory management(AIM),which uses only the current sales data to make an ordering decision for the next period.This algorithm can generate an order quantity for the next period only with the current period data.However,there is also a shortcoming:historical demand information is not fully utilized,resulting in data waste.To overcome the shortcoming,we design an algorithm that uses current sales data and historical sales information summarized in statistics.Our approach takes the advantages of online learning and offline learning.Through extensive experiments,we show that our algorithm leads to better results in general.In this thesis,we simulate 200 instances for each of seven distributions to evaluate the average cost,the convergence rate,and the standard deviation of the average cost of the algorithm.For each distribution,we compare NAIM with AIM,CAVE,and BURNETAS-SMITH,respectively.The numerical results show that NAIM results in lower average costs and faster convergence rates than AIM does in most cases.When the shortage cost is high,we find that although NAIM may achieve the lowest average costs among the four methods,the standard deviation of the average cost is slightly higher than the ones derived from other methods.