Research On Inventory Management Problem Based On Nonparametric Setting

As an indispensable part of modern enterprise operation,inventory management is closely related to the interests of the enterprise.An effective inventory control strategy can reduce the inventory cost of the enterprise and reduce the loss of sales and goodwill caused by the shortage of goods.On the contrary,an inefficient inventory control strategy will increase the related costs and losses of the enterprise and then affect its survival and development.Inventory management aims to help enterprises make reasonable inventory decisions under the constraints of different conditions and support the production and operation of enterprises.Therefore,keeping the inventory at a reasonable level and building an effective inventory management system is the key for a modern enterprise to develop its core competitiveness and achieve long-term development.With the development of the economy and society,the complexity of the environment faced by enterprises is increasing,which greatly increases the difficulty of operation of enterprises and makes the existing strategic effects of enterprises are greatly challenged.In recent years,with the comprehensive penetration of information technology and the advancement of data mining technology,enterprises have begun to enter the era of big data,which has greatly improved the ability of enterprises to collect information.Data has become a valuable asset for enterprises.Therefore,the research on data-driven inventory management based on the nonparametric setting of demand is in line with reality and has vital practical significance.In inventory management,before unknown demand,decision-makers need to determine the inventory level of commodities and use this to meet subsequent demand.Researches often assume that demand distribution is wholly known in traditional inventory problems.But with the drastic change in the business environment,demand uncertainty increases significantly,so inventory control based on random demand arises at the historic moment.In recent decades,the problem of data-driven inventory management has become a key research direction.Many studies have begun to assume that any information on demand is unknown in advance,and decisions are made only based on realized demand data.This setting is more in line with today’s complex demand environment.Following this kind of research trend,this paper studies the problem of inventory management based on the nonparametric setting of demand from the perspectives of regular and perishable products.First,this paper considers a multi-stage newsvendor problem for the research gap in the multi-period newsvendor problem from the perspective of regular products.Also,the advance purchase discount is considered in this problem.To solve this problem,we utilize the Weak Aggregating Algorithm from computer science and propose an extended data-driven algorithm.Unlike regular products,perishables need to consider the impact of the item’s shelf life on the inventory decision.It can only be used to meet the demand within a fixed or random short period,and it will be discarded beyond this period,such as blood products,etc.This paper considers the situation when the commodity is perishable.This paper proposes two deep reinforcement learning algorithms to solve the problem and explore the characteristics of inventory strategies under different conditions.The main research contents of this paper have the following two parts:First,this paper studies a multi-stage newsvendor problem based on the nonparametric setting.It extends the multi-period newsvendor to a multistage newsvendor problem.The planning horizon of the problem consists of multiple stages,and each stage has multiple periods.At the beginning of a stage,the decision maker(DM)places the advance order for all the periods in the stage;At the beginning of each period in the stage,the DM places a regular order.There is no assumption on the demand,the available information the DM can observe is only the past demands.This paper extends the weak aggregating algorithm with one decision variable,an online learning approach based on the theory of prediction and learning with expert advice,to a twodimensional problem that involves advance ordering decisions in stages and regular ordering decisions nested in each stage.The difficulty of the problem lies in transferring learned knowledge for regular ordering decisions from stage to stage.This paper designs a cross-stage experience transfer to adjust regular ordering decisions,and obtain online ordering solutions for advance order and regular-order.In addition,we derive theoretical guarantees for total gains in one stage and cumulative gains for all stages in the planning horizon,which ensure that the cumulative gains converge to the optimal solution for a sufficiently large horizon.Through numerical studies,we find the online ordering solutions obtained by our algorithm are competitive to those offered by the best experts in hindsight.Finally,we do the sensitivity analysis to illustrate the effectiveness of our algorithm under different parameter values.Second,This paper studies a perishable joint pricing and inventory control problem based on the nonparametric setting.Different from researches on regular products,this paper considers the situation when the commodity is perishable.Inventory management of perishables is more complicated than regular products because it requires managing commodities with different shelf lives and considering the impact of shelf life on inventory decisions.Since the existing research has considered the simple inventory management of perishables,this paper takes pricing,lead time,and fixed ordering cost into consideration.These factors have not been considered simultaneously in the existing research because they may make the current solution lose its original effect and fall into the curse of dimension.In this problem,the DM doesn’t know any information about the demand in advance,and the information that can be obtained comes from the feedback of the environment.This paper also considers two unmet demand handling principles: backlogging and lost-sales cases.This paper sets a rational model based on deep reinforcement learning from machine learning and proposes two deep reinforcement learning methods to solve the problem.To show the effectiveness of our proposed algorithms,this paper sets a theoretical upper bound on profits and compares proposed algorithms with other methods.Numerical studies verify the effectiveness of the proposed algorithms and the problem of the “curse of dimension”is circumvented.This paper also discusses the different properties of policies learned under other conditions,which further enriches the research on the properties of optimal policy under complex situations.Based on the nonparametric setting,theoretically,this paper first studies a multi-stage newsvendor problem with advance purchase discount,which riches the research on multi-period newsvendor problem.This paper also extends the weak aggregating algorithm with one decision variable to a two-dimensional problem that involves advance ordering decisions in stages and regular ordering decisions nested in each stage.The new algorithm gives the explicit ordering policy for the regular-order and advance-order and theoretical guarantees for the cumulative gains on one stage and the whole horizon.In addition,the theoretical convergence rate to the optimal policy is offered,which enriches the theoretical results of the study.When studying the complex perishable goods inventory management problem,this paper considers many existing inventory decision-making factors,which enriches the perishable goods inventory management research.In addition,the traditional dynamic programming method is prone to fall into the dimension problem when dealing with such a complex problem.In this paper,the deep reinforcement learning algorithm is used to solve the problem and obtain good results,which verifies the application prospect of the deep reinforcement learning algorithm in the inventory control problem of perishable joint pricing.From the application level,the influencing factors of inventory decision and inventory problem setting considered in this paper have been corresponding to scenes in real life,and experiments have verified the effectiveness of the proposed method.Therefore,the proposed method has potential practical application value and can be used to guide the solution of practical problems.