收益管理系统多分布无约束估计方法仿真比较

Simulation Comparison of Unconstraining Methods Based on Multi-Distribution of Demand in Revenue Management Systems

在优化收益管理系统问题的研究中,无约束估计是收益管理系统中"无约束需求预测"成功实施的关键步骤,对其准确性和有效性进行正确评价的难点技术是建立符合收益管理实践的仿真模型。已有研究仅对采用正态分布的无约束估计方法进行了仿真比较,这不完全符合顾客"无约束需求"数据的多分布特征。另采用EMSR-a模型建立收益管理多分布无约束估计仿真比较模型,对采用正态、伽玛和对数正态分布的Spill模型和EM算法在多分布需求环境中的无约束估计过程进行比较和评价。试验结果验证了仿真比较模型的有效性,证明当需求趋于不稳定时,EM算法性能明显优于Spill模型,需求变异程度是决定EM算法分布假设的主要因素。

Unconstraining estimation is the crucial step in the successful implementation of ＂unconstrained demand forecasting＂ in revenue management systems. The difficult technology to achieve the correct evaluation of its accuracy and effectiveness is to develop a simulation model that meets the practice of revenue management. Existing researches are only focused on the comparative studies of unconstraining methods based on normal distribution, but those are not suitable for the muhi - distribution characteristics of unconstrained demand data. For this problem, a simulation comparison model of unconstraining estimation for multi - distribution of demand in revenue management systems was established based on EMSR- a model. The Spill models and EM algorithms based on Normal, Gamma and Lognormal distribution were compared and evaluated under the multi - distribution demand environments through simulation experiments using practical airline passenger demand data. Simulation results show the effectiveness of the proposed simulation model, and indicate that when demand becomes more unstable, the performance of EM algo- rithms are much better than those of Spill models, the degree of variation in demand is the main factor that determines the distribution assumptions of the EM algorithm.