基于马尔科夫链的网络决策分析方法

在继承传统层次分析/网络分析方法的基础上提出一种新的决策方法——基于有限状态齐次马尔科夫链的网络决策方法MC-ANP.该方法使用有向图定义决策准则及准则之间的支配关系.通过对准则的两两比较量化支配关系,用马氏链的状态转移描述支配关系.该方法突出了准则支配关系的合成过程,提供了两种合成模型(其中积合成模型可以彻底解决传统层次分析中出现的逆序问题),指出传统网络分析中无条件使用Cesaro平均极限求解存在的问题.新方法将决策问题分为两类:有方案的决策问题,解是对方案的排序,求解过程就是求马氏链状态转移概率方阵的属于特征根1的特定的左特征向量;无方案决策问题,解是对准则的排序,解是马氏链状态转移概率方阵的属于特征根1的右特征向量.     

面向多极值质量特性的全局式序贯性实验设计方法

针对多极值质量特性的全局性建模和参数优化问题, 为降低实验设计样本量, 提高模型预测性能, 提出一种全局式序贯性设计方法. 首先在一定的初始实验设计方式下, 建立过程粗略的SVR模型; 而后根据模型中支持向量的分布, 在支持向量样本各维度的45度角或轴向方向同步地增加实验点, 再拟合过程新的SVR模型; 如此迭代进行, 当模型精度达到要求或样本量达到上限时终止序贯性设计. 仿真与实证研究表明, 该方法能够在可行域全局的范围内将实验点合理地分配在质量特性的多个极值附近, 避免了传统的单路径式序贯性设计只能发现单个极值的不足, 充分提高了实验效率; 与均匀空间网格设计、 拉丁超立方设计和均匀设计相比, 在样本量接近的情况下, 基于全局式序贯性设计的SVR 模型的预测均方误差至少降低了30%; 而在预测误差较为接近的情况下, 全局式序贯性设计的样本量至少降低了12%.     

具有对偶极点的不稳定过程的鲁棒控制

为了进一步完善不稳定系统的鲁棒控制器设计方法,将具有对偶极点的不稳定过程分成纯不稳定过程与一般的稳定过程两部分,对于纯不稳定过程采用先成形再镜像映射的方法,而对于一般的稳定过程则基于极点对消原则直接设计控制器,最后用闭环增益成形算法求解鲁棒控制器。将所设计的鲁棒控制器应用于磁浮列车的悬浮控制中,从仿真结果可以看出,控制效果可以达到小超调(<8%)、无静差、调节时间快(<0.1 s)、对模型摄动具有鲁棒稳定性。     

双渠道订货条件下基于预测更新过程的报童模型研究

零售商可以在整个计划时段内持续不断的收集市场需求信息,对市场需求进行预测.离销售季节开始越近,市场需求预测越准确,但采购成本会较高.零售商有两次订货机会.在计划时段开始通过正常渠道订货,单位采购成本最低.在计划时段内零售商还有一次紧急渠道订货机会,可以动态地根据市场需求预测更新结果决策是否订货.当市场需求预测更新过程为MMFE (martingale model of forecast evolution),且在各时点存在不同固定订货成本条件下,证明正常渠道和紧急渠道最优订货策略分别为状态相关的basestock和(s,S)策略,最优订货点与当前时点市场需求预测值线性相关.建立了最优订货策略的求解算法.分析了正常渠道和紧急渠道订货对库存管理的不同意义,结合敏感性分析说明零售商如何在正常渠道和紧急渠道,以及市场需求信息和采购成本间进行权衡.通过大量数值算例发现双渠道订货充分地进行了权衡.     

A production optimization model of supply-driven chain with quality uncertainty

Supply-driven chain’s production is different from traditional demand-driven production because its supplies must guide the full production flow toward the markets and respond actively to customer demand. According to the control theory, a novel multi-variable operation model of supply-driven chain is discussed here, integrating suppliers, manufacturers, distributors and market demands. Especially the coordination problem between suppliers and manufacturers is discussed where suppliers play more important role than manufacturers. Because defect is common in real production system, the production operation of supply-driven chain with imperfect quality is described on the basis of fuzzy set to model the ambiguity of quality and to provide appropriate supply coordination mechanism. In a designed numerical example, it is apparent that both response and robustness performances of supply-driven production system on demand with imperfect quality are improved by a fuzzy proportional-integral- differential regulator. The proposed model may apply to similar productions with imperfect quality.     

基于NSGA-Ⅱ的非平衡制造数据关键质量特性识别

针对非平衡产品制造数据关键质量特性(critical to quality characteristics,CTQs)识别,提出基于NSGA-Ⅱ的特征选择算法.首先,在分类错误率和特征子集大小基础上,针对数据非平衡性,引入第Ⅱ类错误率度量质量特性子集的重要性.接着,应用多目标进化算法NSGA-Ⅱ最小化以上三个度量标准,得到非支配解集.最后,引入理想点法从非支配解集中选择最佳调和解,得到CTQ集.算例结果表明,所提算法能够得到较高分类精度,同时有效降低第Ⅱ类错误率与CTQ集大小,说明了算法的有效性.     

FTC of hidden Markov process with application to resource allocation in air operation

This paper investigates the feedback control of hidden Markov process(HMP) in the face of loss of some observation processes.The control action facilitates or impedes some particular transitions from an inferred current state in the attempt to maximize the probability that the HMP is driven to a desirable absorbing state.This control problem is motivated by the need for judicious resource allocation to win an air operation involving two opposing forces.The effectiveness of a receding horizon control scheme based on the inferred discrete state is examined.Tolerance to loss of sensors that help determine the state of the air operation is achieved through a decentralized scheme that estimates a continuous state from measurements of linear models with additive noise.The discrete state of the HMP is identified using three well-known detection schemes.The sub-optimal control policy based on the detected state is implemented on-line in a closed-loop,where the air operation is simulated as a stochastic process with SimEvents,and the measurement process is simulated for a range of single sensor loss rates.