Research On Data-driven Inventory Management—newsvendor Problem

Inventory management is an indispensable and important part in the process of enterprise operation management,which plays a decisive role in enterprise production and operation activities.Proper inventory can reduce inventory cost without affecting normal production and customer service,while excessive inventory will cause the burden of carrying cost.Therefore,how to plan and purposefully control the inventory in a “reasonable” range,to establish a set of effective inventory management core technology,is the key to improve the core competitiveness and achieve long-term development of enterprises.On the other hand,with the advancement of the new generation of technologies represented by mobile Internet,cloud computing and big data,the mining and utilization of data value has become an important purpose of informatization.Therefore,with the intensification of enterprise competition and the increasing uncertainty of customer demand,the sustainable development of enterprises cannot be separated from data-driven decision,and the datadriven inventory management method meets the needs of the development of the times.The newsvendor problem is the basis of operations and supply chains models in inventory management.Before the real demand is known,the decision maker needs to determine the inventory of goods to effectively meet the real demand.In the newsvendor problem with random demand and unknown distribution,differ from the methods with parametric information,this paper proposes a robust optimization method based on data-driven nonparametric information from the perspective of distributionally robust optimization method,aiming at the problem that the existing methods do not make enough use of nonparametric information.In addition,for the newsvendor problem with demand distribution of one or more unknown parameters,this paper studies and explores the operational statistics from the perspective of Bayesian analysis.The main research contents of this paper are as follows:First,the robust inventory management method based on nonparametric information and its application.Under the premise that the demand distribution is unknown and no assumptions are made,an innovative data-driven method is proposed in this paper,which uses the non-parametric characteristics of the distribution,such as monotonicity and concavity,to construct the ambiguity set of the distribution.Under the fixed partitioning method,the constructed distribution ambiguity set contains all distributions that share the desired nonparametric characteristics with the real distribution,and the protection curve is used as a proxy for the worst-case distribution in the ambiguity set to approximate the real demand density curve.In this paper,the value of nonparametric information is verified theoretically and experimentally,and the application value of the proposed method in inventory management is further illustrated by comparing with other methods.Second,the robust inventory management method based on adaptive adjusted nonparametric and its application.Based on the nonparametric robust optimization method under fixed partitioning,the biases of data-driven nonparametric estimation are further considered.By analyzing the stability and convergence of the nonparametric estimation when the new data is updated,a criterion for judging the unbiasedness of nonparametric features in an interval with different data size is established,and an adaptive partitioning nonparametric robust method is proposed.This method can adjust the distributional ambiguity set according to the data input,so that the distributional ambiguity set can ensure that the real demand distribution is included in this constructed distributional ambiguity set with a certain probability.In addition,a data size threshold is proposed to effectively balance the advantages of empirical methods and nonparametric robust methods,and is illustrated in experiments.Thirdly,the parametric inventory management method based on Bayesian analysis and its application.In view of the newsvendor problem with known demand distribution but unknown parameters,considering the operational statistic method,in which optimal decision is directly estimated from data,compared with the traditional separate estimation and optimization method,can bring improvement in expect profit.But in most cases such as distribution has more unknown parameters,the closed-form operational statistics is hard to derive.Therefore,we propose a general data-driven method under the Bayesian framework.For the parametric distribution in newsvendor model with sufficient statistics,the polynomial function related to the closed-form operational statistics is obtained by fitting,and it is verified and applied in exponential distribution and pareto distribution.We propose a series of data-driven inventory management methods for the random demand inventory management newsvendor problem.At the theoretical level,we propose a robust inventory management method based on nonparametric information,which is different from the previous parametric methods and nonparametric methods based on statistical distance.The distribution ambiguity set is segmented by nonparametric features such as monotonicity and concavity,and the constructed protection curve is used to make inventory decision.In addition,in view of the newsvendor problem with parametric distribution,a polynomial fitting method based on Bayesian framework is proposed,which provides theoretical support and guidance for solving the closed-form operational statistics of general distribution or with multiple unknown parameters.At the application level,the proposed nonparametric robust inventory management method can be effectively applied to newsvendor continuous and discrete problems.By using the real data from an open platform and sample data of arbitrary distribution,we compare our method with existing parametric method and nonparametric method,and find that in some cases our method can significantly improve the profits,profit rate and stability.In addition,by verifying the value of nonparametric information and proposing a robust optimization model by combining parametric information and nonparametric information,it provides suggestions and guidance for decision makers to use which information and model in practical applications.