Linear regression and sensitivity analysis in nuclear reactor design. (November 2015)

• Record Type: Journal Article
• Authors: Kumar, Akansha; Tsvetkov, Pavel V.; McClarren, Ryan G.
• Is Part Of: Annals of nuclear energy. Volume 85(2015:Nov.)
• Page Start: 798
• Page End: 811
• Publication Date: 2015-11
• Subjects: Predictive -- Optimization -- Reactor -- Regression -- Sensitivity -- Fast; Nuclear energy -- Periodicals; Nuclear engineering -- Periodicals; 621.4805
• DOI: 10.1016/j.anucene.2015.06.037
• Languages: English
• ISSNs: 0306-4549

Abstract

Highlights: Presented a benchmark for the applicability of linear regression to complex systems. Applied linear regression to a nuclear reactor power system. Performed neutronics, thermal–hydraulics, and energy conversion using Brayton's cycle for the design of a GCFBR. Performed detailed sensitivity analysis to a set of parameters in a nuclear reactor power system. Modeled and developed reactor design using MCNP, regression using R, and thermal–hydraulics in Java. Abstract: The paper presents a general strategy applicable for sensitivity analysis (SA), and uncertainity quantification analysis (UA) of parameters related to a nuclear reactor design. This work also validates the use of linear regression (LR) for predictive analysis in a nuclear reactor design. The analysis helps to determine the parameters on which a LR model can be fit for predictive analysis. For those parameters, a regression surface is created based on trial data and predictions are made using this surface. A general strategy of SA to determine and identify the influential parameters those affect the operation of the reactor is mentioned. Identification of design parameters and validation of linearity assumption for the application of LR of reactor design based on a set of tests is performed. The testing methods used to determine the behavior of the parameters can be used as a general strategy for UA, and SA of nuclear reactor models, and thermal hydraulics calculations. A design of a gas cooled fast breeder reactor (GCFBR), with thermal–hydraulics, and energy transfer has been used for the demonstration of this method. MCNP6 is used to simulate the GCFBR design, and perform the necessary criticality calculations. Java is used to build and run input samples, and to extract data from the output files of MCNP6, and R is used to perform regression analysis and other multivariate variance, and analysis of the collinearity of data.