Machine learning-based 2D model optimization design for fusion target pellets

Abstract

This paper proposes a machine learning-based method for the optimal design of a two-dimensional model of a fusion target pellet. The optimized design of the 2D model of the fusion target pill based on machine learning includes: parameterization of the fusion target pill, mesh division, construction of a database, data dimensionality reduction, construction of a neural network, and genetic algorithm optimization. On the basis of the original 2D model grid of the fusion target, the H-type grid is kept unchanged, only the O-type grid is replaced, and the total number of grid points is kept unchanged, so that the 2D model of the modified fusion target with any modeling parameter can quickly extract the physical parameters at the specified position in the flow field, with better geometrical adaptability, and improve the efficiency of data processing effectively. By combining the principal component analysis method and artificial neural network method to achieve the reconstruction of the flow field of the two-dimensional model of the fusion target, under the premise of guaranteeing the accuracy requirements, reducing the dimension and compression of the computational volume, thus reducing the time-consumption and saving the computational resources, realizing the rapid evaluation of the radiation hydrodynamic performance of the two-dimensional model of the fusion target and the rapid search for the optimization of the two-dimensional model of the fusion target, and providing a new idea for the high-efficient design of the two-dimensional model of the high-performance fusion target.