High-dimensional multi-objective shielding optimization method based on multi-parameter shielding calculation agent model

•A multi-parameter shielding calculation agent model achieves high precision and efficient computation in complex shielding, unlike traditional neural networks.•The improved NSGAIII algorithm (k_NSGAIII) with an knee point strategy optimizes high-dimensional multi-objective problems, suitable for co...

Full description

Saved in:
Bibliographic Details
Published in:Annals of nuclear energy Vol. 214; p. 111182
Main Authors: Gui, Long, Song, Yingming, Yuan, Weiwei, Xia, Yue
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.05.2025
Subjects:
High-dimensional multi-objective
k_NSGAIII
Neural network
Shielding calculation agent
Shielding optimization
Swish-NNTA
Neural network
Shielding optimization
Shielding calculation agent
High-dimensional multi-objective
Swish-NNTA
k_NSGAIII
ISSN:0306-4549
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A multi-parameter shielding calculation agent model achieves high precision and efficient computation in complex shielding, unlike traditional neural networks.•The improved NSGAIII algorithm (k_NSGAIII) with an knee point strategy optimizes high-dimensional multi-objective problems, suitable for complex reactor shielding structure optimization.•High-dimensional multi-objective optimization is conducted on reactor shielding benchmark models with 7 objectives and 12 variables, validating the method's effectiveness and k_NSGAIII's advantages. The design of complex reactor shielding structures requires consideration of neutron and photon radiation levels in various regions, as well as trade-offs in weight, volume, and cost, leading to a substantial increase in shielding calculation parameters and optimization objectives. This study introduces a high-dimensional multi-objective shielding optimization method based on a multi-parameter shielding calculation surrogate model, with significant improvements to the FCNN-NSGAIII shielding optimization method which integrates a neural network with a genetic algorithm. For the optimization of complex reactor shielding structures under diverse source item energy spectra, the multi-parameter surrogate model achieves a prediction error reduction of an order of magnitude to 3.65% compared to traditional neural networks. Furthermore, the k_NSGAIII optimization algorithm, enhanced with a knee-point strategy, demonstrates a greater relative set coverage indicator than the NSGAIII algorithm, indicating its ability to identify superior shielding design schemes.
ISSN:0306-4549
DOI:10.1016/j.anucene.2024.111182