Surrogate Machine Learning Model for Multi Objective Simulated Annealing-Based Core Reloading Pattern Optimization

This study explores the integration of artificial neural networks (ANNs) with a simulated annealing (SA) algorithm to optimize nuclear reactor core loading patterns. By employing ANNs to predict critical reactor parameters, such as reactivity, power peaking factor (PPF_max), and cycle length in days, and combining them with the SA algorithm for optimization, the study addresses the multi-objective challenge of enhancing reactor efficiency and safety. The fitness function, defined as the ratio of cycle length to PPF_max, serves as the optimization objective, enabling the SA algorithm to identify loading patterns that maximize operational cycle length and minimize the power peaking factor. The methodology incorporates a high-dimensional design space with factorial complexity and leverages the predictive accuracy of ANNs to guide optimization. Results demonstrate the framework’s ability to improve reactor performance metrics, achieving longer operational cycles and reduced safety constraints. The findings underscore the potential of integrating advanced machine learning and heuristic optimization techniques in reactor design.