Parametric Analysis and Optimization using Multivariate Regression Analysis and Genetic Algorithms

摘要

Parametric analysis and optimization of a specific set of neutronics parameters of a thorium-fueled pressurized heavy water reactor core fuel has been performed using Genetic Algorithms (GA), and a multivariate regression method. This work performs a detailed pin-cell analysis of a seed-blanket configuration, where the seed is composed of natural uranium, and the blanket is composed of thorium. Through this work, we fulfill the following objectives: Firstly, identify, and analyze the behavior of core parameters in the reactor design of breed-and-burn type thorium fuel. Secondly, determine an optimized combination of the pitch-to-diameter ratio (- ratio), and material composition to satisfy a set of neutronics objectives. Finally, explore a novice optimization method using an integration of GA and predictive analysis using a regression model. Regression analysis is done using, recursive partitioning of decision trees (rpart) multivariate regression model. Reactor designs are usually complex and a simulation needs a significantly large amount of time to execute, hence implementation of GA or any other global optimization techniques by itself is not feasible, therefore we present a new method of using rpart in conjunction with GA. Due to using rpart, we do not necessarily need to run the neutronics simulation for all the inputs generated from the GA module rather, run the simulations for a predefined set of inputs, build a regression to the input and the output parameters, and then use this to predict the output parameters for the inputs generated by GA. The rpart model is implemented as a library using R programming language, and the neutronics model, design and analysis has been done using Serpent 1.1.19 Monte Carlo code.