A method for comparing heuristics with applications in computational design of flexible systems

Design;Real options;Value of information;Hybrid energy systems;Risk management;Dynamic programming;Heuristics

Binder, William ; Paredis, Christiaan J. J. Mechanical Engineering Jeter, Sheldon M. Fu, Katherine McGinnis, Leon German, Brian Garcia, Humberto E. ; Paredis, Christiaan J. J.

University:Georgia Institute of Technology

Department:Mechanical Engineering

关键词: Design;    Real options;    Value of information;    Hybrid energy systems;    Risk management;    Dynamic programming;    Heuristics;   

Others  :  https://smartech.gatech.edu/bitstream/1853/58660/1/BINDER-DISSERTATION-2017.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

All designers must make tradeoffs when making a decision. Ideally, a designer makes the decision that maximizes their value. However, selecting the correct alternative may not be obvious. To help make decisions, designers may employ heuristics, rules of thumb that recommend actions. But even selecting the best heuristic may be challenging. To best select a heuristic requires that we be able to compare different heuristics. Currently, there are no formal methods to fairly compare different heuristics. This dissertation proposes a Design Decision Framing Model (DDFM) and method for comparing heuristics to enable designers to make better decisions by identifying more valuable heuristics. To demonstrate the method we focus on one research area that would greatly benefit from better heuristics: the field of flexible design. Flexible design explicitly recognizes that after a particular decision is made, subsequent decisions will be made that influence the value of an artifact. As such, designers should consider the effect of these subsequent decisions to make the best current decision. However, analyzing subsequent decisions can be very challenging. To address this, different heuristics have been suggested, but it is difficult to know which heuristic is best. In all likelihood, different heuristics perform better than others under different conditions. Thus, part of the challenge is in identifying the conditions under which a particular heuristic is most preferred. The DDFM is applied to compare the performance of different flexible-design heuristics. This comparison suggests that the DDFM is useful and can be used with a research method to characterize heuristics. Further, this dissertation proposes a new heuristic for analyzing flexible systems. The new heuristic uses dynamic programming and surrogate modeling to efficiently analyze future decisions, as well as provide insight into the reasons why certain decisions were made. The new heuristic is investigated in a case study of a Hybrid Energy System (HES). HES are long-lived systems subject to large uncertainty, making them particularly challenging to analyze, especially in a flexible design context. The results of this case study suggest that the new heuristic can be advantageous under certain conditions.