基于样本划分的数据驱动报童问题研究OA北大核心CHSSCDCSSCICSTPCD

The advent of the era of big data has brought unprecedented opportunities for enterprise inventory management,but also has raised new challenges to its corresponding decision-making methods.On the one hand,the richness and availability of data in the era of big data has brought rich information to support enterprise's decision-making,such as real-time customer behavior data and product-related data owned by large e-commerce platforms.How to efficiently use big data to support inventory management has become a new growth point for enterprises.On the other hand,the big data method has promoted the transformation of research paradigm in the field of inventory management.Traditional inventory management typically assumes that the demand distribution function is known.However,the real data is usually massive,multi-dimensional and unstructured,which raised new big challenges for statistical methods,and it becomes more difficult to estimate and predict.In the context of big data,the research paradigm of inventory management has also undergone significant changes,from traditional model driven to data-driven decision-making.Data driven decision-making is to skip the link of demand forecasting and directly establishes the relationship between data and decision-making. Efficient use of real-time market data to support intelligent decision-making for inventory management has become the key for enterprises to improve the efficiency of inventory management.This paper studies the classical newsboy problem,that is,newsboy needs to determine the order quantity in each period to meet the random demand.The random demand distribution is unknown,and newsboy has only several periods of historical demand data.The goal of newsboy is to minimize inventory costs,including shortage costs and inventory holding costs.Based on the data sample partition,this paper proposes a new data-driven robust optimization method(DROSP)to solve the newsboy ordering problem.DROSP method divides the demand data samples into different intervals to construct the fuzzy set of random demand distribution,so as to solve the corresponding robust optimization problem.The original problem is difficult to solve.By redefining the cost structure and sample set,the original problem is transformed into a classical robust distributed optimization problem,which can be solved efficiently.This paper also theoretically proves that the solution of DROSP method converges to the optimal solution with the increase of sample size.Numerical experiments compare the efficiency of DROSP and classical SAA methods.Four different types of demand distributions are tested,including multi-peak random distribution,high standard deviation distribution,heavy head distribution and heavy tail distribution.The numerical results show that:1)firstly,the efficiency of both DROSP and SAA methods first increases and then decreases as the service level increases;2)the efficiency of the DROSP method proposed in this paper is obviously better than that of the traditional method SAA,especially when the service level is in the middle range.However,when the service level is small or large,the efficiency of the two is similar.At the same time,under the multi peak random distribution,heavy head distribution and heavy tail distribution,the efficiency of DROSP method is much better than that of traditional SAA method,while the efficiency of the two methods is the same under the high standard deviation distribution;3)in most test examples,the error standard deviation of the DROSP method proposed in this paper is less than that of the SAA method,which means that the robustness of the DROSP method is stronger than that of the SAA method;4)with the increase of sample size,the efficiency of both methods will become better,and the change trend is similar.This paper proposes a new robust optimization method and applies it to data-driven inventory management.The research not only enriches the development of data-driven inventory management theory,but also provides more data-driven inventory algorithms for enterprise practice,and provides more feasible and effective theoretical and methodological support for enterprise intelligent decision-making.