Big Models using Big Data for Simulation-based Design and Operational Optimization

使用大数据进行基于仿真的设计和操作优化的大模型

• 批准号: RGPIN-2018-06589
• 负责人: Srinivasan, Balasubrahmanyan
• 金额: $ 2.04万
• 依托单位: École Polytechnique de Montréal
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=714349
• 关键词: Big; Models; using; Data; Simulation

Simulation-based engineering science is a rapidly growing domain where numerical simulation of complex models are used to make design decisions. However, for the models to represent reality, it is better that they be based on collected data. On the other hand, an emerging area of research activity is the Big Data analytic that uses large volume of data in real-time from a variety of information sources for decision-making. Also, new developments in the modeling field relates to network systems, where a number of sub-models are connected with a given topology. This proposal combines all the above domains, by constructing Big Models from Big Data, which will then be used for simulation-based design and optimal operation. Big Models are clusters of different types of sub-models, derived from Big Data, to represent a given system. Big Models inherit the four Vs of Big Data. The volume aspect of Big Models corresponds to the number of variables used, the variety aspect to the different types of sub-models that co-exist, the velocity aspect to the dynamic or time-varying nature, and the variability aspect to the model uncertainty. It is noted that the construction of Big Models requires a fractured approach in the beginning, followed by an eventual integration. The proposal proposes a systematic methodology for the construction of such a network model. It also develops algorithms for regression with Big Data. The proposal continues with the two aspects of Process Systems Engineering, i.e., the design and operation. Design decisions are made either using a sensitivity analysis, a mixed-integer programing along with a heuristic simplification. Safety and flexibility are taken into consideration. Operational decisions are taken based on a combination of numerical optimization and real-time Big Data. For the latter, a multi-unit approach is proposed, where a Big Model runs in parallel to the system and the difference between data simulated by the Big Model and the obtained Big Data is used for adaptation. The proposed methodology will be applied to two case-studies, (i) the aircraft environmental control and (ii) waste water treatment using microbial fuel cells. Both of them have heterogeneous subsystems, and each of the sub-systems has a completely different modeling methodology. The data are also very varied, from direct to indirect measurements. The design aspect will be tested on the aircraft problem, while operational optimization will be carried out on the waste water treatment case.

基于仿真的工程科学是一个快速发展的领域，其中复杂模型的数值仿真用于做出设计决策。然而，为了使模型能够代表现实，最好是基于收集的数据。另一方面，一个新兴的研究活动领域是大数据分析，它实时使用来自各种信息源的大量数据进行决策。此外，建模领域的新发展涉及网络系统，其中许多子模型与给定拓扑结构连接。该提案结合了上述所有领域，通过从大数据构建大模型，然后将其用于基于仿真的设计和优化操作。 大模型是从大数据中派生的不同类型的子模型的集群，以表示给定的系统。大模型继承了大数据的四个V。大模型的体积方面对应于所使用的变量的数量，多样性方面对应于共存的不同类型的子模型，速度方面对应于动态或时变性质，可变性方面对应于模型的不确定性。值得注意的是，大模型的构建需要在开始时采用断裂的方法，然后进行最终的整合。 该提案提出了一个系统的方法来构建这样一个网络模型。它还开发了大数据回归算法。该建议继续过程系统工程的两个方面，即，设计和操作。设计决策是使用灵敏度分析，混合整数规划沿着与启发式简化。考虑了安全性和灵活性。运营决策基于数值优化和实时大数据的组合。对于后者，提出了一种多单元方法，其中大模型与系统并行运行，并且大模型模拟的数据与获得的大数据之间的差异用于适应。 所提出的方法将应用于两个案例研究，（i）飞机环境控制和（ii）废水处理使用微生物燃料电池。它们都有异构的子系统，并且每个子系统都有完全不同的建模方法。从直接测量到间接测量，数据也非常多样化。设计方面将在飞机问题上进行测试，而操作优化将在废水处理的情况下进行。