A Gradient-based Sensitivity Analysis Method for Complex Systems

Abstract

Most sensitivity analysis methods from literature impose specific experiment plans (design-of-experiments). Moreover, the size of the experiment plan, which is in conjunction with the number of system evaluations, increases with the number of factors that may affect the systems' behavior. This paper introduces a gradient-based global sensitivity analysis method which overcomes these limitations. First, its performance is compared against six sensitivity analysis methods on sets of polynomial test functions. The comparison is carried out by means of the number of system evaluations implied and the reported factor ranking list. The proposed method proved to have comparable accuracy to the best of the six known methods- the EFAST method- with the advantage of a lower number of system evaluations. These two methods are further applied on an electronic system, an E-Bike application. In this case, the proposed method employs the verification plan of the EFAST method, as well as a standard Monte Carlo experiment plan with about one third of the system evaluations of the EFAST. Even with a much lower number of system evaluations, the proposed method yields the same result as the EFAST method in terms of the factor ranking list.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 10 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.