压水堆换料优化中Bagging-NN-ACO联合算法应用研究

Application Research of Bagging-NN-ACO Joint Algorithm in PWR Refueling Scheme Optimization

摘要: 在压水反应堆停堆换料中，换料方案直接决定了反应堆运行的安全性与经济性，其设计涉及热工水力、中子物理、反应堆控制等复杂问题，导致从大量堆芯换料方案中筛选最优方案费时费力。本文提出Bagging-NN-ACO联合算法，即利用集成的人工神经网络(Bagging-NN)实现堆芯参数快速、精确计算，利用蚁群算法实现大量换料方案的优选。在压水堆堆芯参数预测问题中，Bagging-NN方法通过“全局随机抽样 + 重点局部随机抽样”方法及组合多个神经网络模型，可以使有效增值因子参数的预测误差在0.1%以内，其他因子也在1.55%以内，较单一神经网络模型拥有更高的预测精度。在燃料组件排布优化中，使用具有自组织优化能力的蚁群算法，以堆芯关键参数为目标进行迭代进化，联合算法实现了换料优化问题的全流程智能计算，能够使堆芯目标参数优化提升2.29%~3.72%。

Abstract: In pressurized water reactor shutdown refueling, refueling scheme directly determines the safety and economy of reactor operation, and its design involves complex problems such as thermal hydraulic, neutron physics, reactor control, etc., which leads to time and effort to select the optimal scheme from a large number of core refueling schemes. In this paper, the Bagging-NN-ACO joint algorithm is proposed, namely, the integrated artificial neural network (Bagging-NN) is used to achieve fast and accurate calculation of core parameters, and the ant colony algorithm is used to optimize a large number of refueling schemes. In the prediction of pressurized water reactor core parameters, the Bagging-NN method, through the method of “global random sampling + key local random sampling” and the combination of several neural network models, can make the prediction error of effective value-added factor parameters within 0.1% and other factors within 1.55%, which has a higher prediction accuracy than the single neural network model. In the optimization of fuel assembly arrangement, the ant colony algorithm with self-organizing optimization capability is used to carry out iterative evolution with core key parameters as the target. The combined algo-rithm realizes the intelligent calculation of the whole process of refueling optimization problem, which can improve the core target parameter optimization by 2.29%~3.72%.