Abstract Safety and lightweight optimization are significant in developing high-performance compact nuclear reactors. We proposed the reactor radiation shielding optimization method by coupling genetic algorithm and neural network. The engineering design is a multi-objective and multi-constrained complex optimization problem. Thus, we improved it by non-dominated sorting genetic algorithm (NSGA-II) and mini-batch stochastic gradient descent (MSGD), and proposed an adaptive mutation rate operator to improve the global searching capability. The multi-objective optimization models were constructed to verify the feasibility of the method. The optimization results proved it can find the Pareto-optimal front of reactor shielding designs precisely and its calculation speed is two orders of magnitude faster than the pure Monte Carlo (MC) method. By contrast with MC, the method can reduce the calculation time cost and have similar computational accuracy, which demonstrated the applicability and effectiveness of the method in the reactor shielding design applications.

中文翻译：

---

核反应堆辐射屏蔽快速智能多目标优化方法研究

摘要 安全性和轻量化优化对于开发高性能紧凑型核反应堆具有重要意义。提出了遗传算法与神经网络耦合的反应堆辐射屏蔽优化方法。工程设计是一个多目标、多约束的复杂优化问题。因此，我们通过非支配排序遗传算法（NSGA-II）和小批量随机梯度下降（MSGD）对其进行了改进，并提出了一种自适应变异率算子来提高全局搜索能力。构建了多目标优化模型，验证了该方法的可行性。优化结果证明它可以精确地找到反应堆屏蔽设计的帕累托最优前沿，其计算速度比纯蒙特卡罗（MC）方法快两个数量级。