An Evidence Theory Based Decision Method for the Variable Sensitivity of Multi-Output Systems

Abstract

Sensitivity analysis is commonly used to identify key parameters of a system and gain insight into the effect of variables on system outputs. For systems with multiple outputs, traditional sensitivity analysis methods are typically carried out for each system response individually. However, obtaining the importance of variables from the system-level view still needs objective decision support. To address this issue, this study proposes a multi-output system sensitivity decision-making method based on evidence theory. The proposed method first conducts surrogate model-based Sobol sensitivity analysis for each output of the system. Subsequently, the global sensitivity information for each output is used as evidence to construct a basic probability assignment, according to a rule defined for determining basic probability assignment. Finally, a synthesis rule is applied to calculate comprehensive sensitivity information. The effectiveness of the proposed method is demonstrated through validation with three numerical examples and one engineering case. This method provides more objective and rational decision support for sensitivity analysis of multi-output systems, offering significant potential benefits in parametric studies of complex systems.