Uncertainty Analysis Based on Sensitivities Generated Using Automatic Differentiation
The objective is to determine confidence limits for the outputs of a mathematical model of a physical system that consists of many interacting computer codes. Each code has many modules that receive inputs, write outputs, and depend on parameters. Several of the outputs of the system of codes can be compared to sensor measurements. The outputs of the system are uncertain because the inputs and parameters of the system are uncertain. The method uses sensitivities to propagate uncertainties from inputs to outputs through the complex chain of modules. Furthermore, the method consistently combines sensor measurements with model outputs to simultaneously obtain best estimates for model parameters and reduce uncertainties in model outputs. The method was applied to a test case where ADIFOR2 was used to calculate sensitivities for the radiation transport code MODTRAN.
KeywordsCovariance Matrix Probability Density Function Uncertainty Analysis Sensor Measurement Automatic Differentiation
Unable to display preview. Download preview PDF.
- 1.Myers, R., Response Surface Methodology, Edwards Bros. (1976).Google Scholar
- 2.Kosko, B., Fuzzy Engineering, Prentice Hall (1996).Google Scholar
- 4.Bishop, M., Neural Networks for Pattern Recognition, Oxford University Press (1997).Google Scholar
- 6.Barhen, J. et al., “Uncertainty Analysis of Time-Dependent Nonlinear Systems”, Nucl. Sci. Eng., 81, 23–44 (1982).Google Scholar
- 8.Toomarian, N. and J. Barhen, “Neural Network Training by Integration of Adjoint Systems of Equations Forward in Time”, U.S. Patent No. 5,930,781 (July 27, 1999).Google Scholar
- 9.Griewank, A., Evaluating derivatives: principles and techniques of algorithmic differentiation, Society for Industrial and Applied Mathematics, Philadelphia (2000).Google Scholar