Combining Simulation-Decomposition, Simulation-Optimization, and Modelling-to-Generate-Alternatives for Planning Under Uncertainty

最近，一种新的探索性可视化方法，模拟分解（SimDec），已被引入，通过增强输入和模拟输出的多变量组合之间的因果关系的解释能力，扩展蒙特卡罗分析。SimDec可以揭示以前无法识别的输入和输出的多变量组合之间的因果关系。SimDec方法可推广到任何Monte Carlo模型，额外的计算开销可以忽略不计，因此，可以很容易地用于采用模拟模型的任何分析。 拟议的研究将扩展SimDec方法的探索性数据的方法，输入/输出识别，分区，场景创建和增强的可视化技术，并调查不同的应用程序在大规模的规划设置。此外，使用方差分解，以前的蒙特卡罗方法计算指数，以评估个别参数的敏感性和它们对输出的相互作用。所谓的Sobol分析确定每个输入参数对模型输出的贡献的一阶和总效应灵敏度指数。沿着这些路线，拟议的研究将探讨如何创建基于SimDec的一阶，高阶和相互作用的灵敏度指数-分析和数值-不仅可以应用于SimDec应用程序，但也可扩展到任何蒙特卡洛模型。仿真优化（SO）提供了一种优化方法，它包含了以概率分布表示的不确定性。在SO中，所有未知的目标，约束和参数都被仿真模型所取代，其中决策变量提供了每个仿真实验运行的设置。高效的优化组件通过可行域引导解决方案探索，仅执行有限数量的模拟。感兴趣的是连接到一个新的，混合SimDec-SO方法的SimDec与SO的优化功能的视觉分析方面，并将其应用到建模生成的替代品。许多“真实的世界”的应用，这SimDec-SO方法将被认为是广泛的测试将确定其适用性在许多不同的规划环境。为了衡量和展示效率的提高，我将继续使用来自几个早期案例研究的数据，这些案例研究在城市固体废物规划、逆向物流、延伸供应链、医疗保健、食品加工、农业和环境可持续性等不同环境中使用了这种技术，并研究了其他感兴趣的领域。这些应用程序都具有相当大的公共重要性。 这项研究的大部分将涉及广泛的数学和计算测试，而某些应用将需要现场访问其他适当的工业和应用环境。

Recently, a new exploratory visualization approach, simulation decomposition (SimDec), has been introduced that extends Monte Carlo analysis by enhancing the explanatory power of the cause-effect relationships between multi-variable combinations of inputs and the simulated outputs. SimDec can reveal previously unidentifiable cause-and-effect connections between multi-variable combinations of inputs on the outputs. A SimDec approach is generalizable to any Monte Carlo model with negligible additional computational overhead and, hence, can be readily used in any analyses that employ simulation models. The proposed research will extend the SimDec method with respect to exploratory data approaches, input/output identification, partitioning, scenario creation, and enhanced visualization techniques, and investigate different applications in large-scale planning settings. Furthermore, using variance decomposition, prior Monte Carlo approaches have calculated indices to evaluate the sensitivities of individual parameters and their interactions on outputs. So-called Sobol analysis determines first-order and total-effect sensitivity indices for the contributions of each input parameter to model output. Along these lines, the proposed study will examine ways to create SimDec-based first-, higher-order, and interaction sensitivity indices - both analytically and numerically - that can be applied not only to SimDec applications, but are also extendable to any Monte Carlo model. Simulation-optimization (SO) provides an optimization approach that incorporates uncertainties expressed as probability distributions. In SO all unknown objectives, constraints, and parameters are replaced by simulation models in which the decision variables provide the settings under which each simulation experiment is run. An efficient optimization component guides the solution exploration through the feasible domain performing only a limited number of simulations. Of interest is to concatenate the visual analytic facets of SimDec with the optimization features of SO into a novel, hybrid SimDec-SO method and to apply it to modelling-to-generate-alternatives. Numerous "real world" applications of this SimDec-SO approach will be considered and extensive testing will ascertain its applicability in numerous disparate planning contexts. In efforts to gauge and demonstrate the efficiency improvements, I will continue with data from several earlier case studies that have used this technique in such diverse settings as municipal solid waste planning, reverse logistics, extended supply chains, health care, food processing, agriculture, and environmental sustainability - as well as examining other areas of interest. These applications all possess considerable public importance. Most of this research will involve extensive mathematical and computational testing, while certain applications will necessitate on-site visits to other appropriate industrial and application environments.