Research on reactor refueling optimization using KAADPN integrated probability distribution guided heuristic algorithm

Abstract

This study addresses the refueling optimization problem for reactors, selecting the effective multiplication factor as the metric for evaluating loading schemes. The Characteristic Statistical Simulated Annealing and Characteristic Statistical Genetic Algorithm are proposed, which significantly enhance the exploration of the solution space and improve the global search capability. The Kolmogorov-Arnold Attention Dual-Path Network (KAADPN) is introduced, combining the function modeling ability of KAN with the global feature capture of the self-attention mechanism. This significantly improves the model’s prediction accuracy while enhancing its computational efficiency. By establishing a surrogate model for core physics calculations and integrating it with optimization algorithms, pseudo-equilibrium optimization analysis is conducted. The effectiveness of the algorithms is compared through single-cycle optimization case studies, and preliminary no-shuffling optimization verification is performed, resulting in ideal core fuel loading schemes. This validates the feasibility of the method and provides a new tool for efficiently addressing the refueling optimization problem.