Point Kernel Modification Including Support Vector Regression Neutron Buildup Factor Models

Abstract

This work presents the results of radiation shielding calculations using modified point kernel code QAD-CGGP. The modification includes a new approach to neutron buildup factor estimations based on machine learning technique called Support vector regression (SVR). SVR neutron buildup factor models for common shielding materials are developed and built into the QAD-CGGP. The development of the models consisted of acquiring the data to be used for learning the model, optimizing the SVR parameters, and application of active learning methods for improving the learning process. The modified code is tested, and the results are compared with the MCNP6 results.