基于PCA的气冷微型堆堆芯在线监测方法研究

Research on Online Monitoring Method for Gas-cooled Micro Reactor Based on PCA

先进微型堆的研发面临仅依靠少量堆外探测器信号实现堆芯在线监测的挑战。本研究提出一种通过主成分分析和堆外探测器空间响应函数进行堆芯功率重构和棒位预测的方法,并在气冷微型堆上初步验证其可行性。研究结果表明,通过核设计程序构建不同工况下堆芯节块功率分布数据库,可以选择少于堆外探测器信号个数的主成分实现堆芯功率分布的降维,且样本恢复精度在2.5%以内;为保证主成分的适用性,应使功率分布数据库包含尽量全面的工况因素;堆外探测器响应函数受其与堆芯节块的相对位置影响显著,轴向探测器响应受前后反射层厚度影响明显,径向探测器响应与全堆燃料组件列均有映射关系;节块级功率重构可基于主成分降维和堆外探测器信号通过最小二乘法实现,重构误差在2.9%以内;轴向芯块级功率重构可基于重构的节块级功率通过多项式拟合实现,最大功率因子拟合误差在0.2%以内;棒位预测可通过构建棒位与功率在主成分上的投影系数之间函数映射模型和非线性最小二乘法拟合实现,棒位预测偏差在0.5 cm以内。该堆芯在线监测方法功率重构、棒位预测精度高,不受堆型和工况限制,可应用在各种先进微堆研发中。

The online monitoring has been a main challenge for the development of advanced micro reactors due to the constrained number of ex-core detectors that can be deployed.A method for core power reconstruction and control rod position prediction was proposed in this study to tackle this problem.It was based on the principal component analysis(PCA)of the core power distributions at various state conditions,with the ex-core detector response functions(DRFs)mapping the core power distribution to the ex-core detector counting.The application of the method on a gas-cooled micro reactor was reported in this paper.The nodal-level power distribution database covering different operating conditions has been constructed using the nuclear design code,followed by the PCA that gives the principal components(PCs).The reduced order was achieved by transforming the determination of the nodal power to determination of the coefficients of the PCs.It is confirmed that PCs of a number less than the number of ex-core detectors can produce the nodal power recovery accuracy of within 2.5%,as long as the power distribution database covers enough range of the core operation conditions.On the other hand,the DRFs show strong spatial dependence,which implies tight coupling of the core power and the ex-core detector counting.Specifically,the axial DRFs are significantly affected by the thickness of the front and rear reflector,and the radial DRFs confirms that no fuel assembly contributes a negligible fraction.Overall,the nodal-level core power can be reconstructed from the ex-core detector signals,with reconstruction errors within 2.9%.The axial pelletlevel power distribution of a fuel assembly can be reconstructed through polynomial fitting with the axial nodal-level power distribution,which gives the accurate peak power factor of error within 0.2%.Furthermore,the prediction of the control rod position can be achieved by constructing a function approximately mapping the coefficients of PCs to the rod position,with the prediction error of within 0.5 cm.This online monitoring method has high accuracy for power reconstruction and control rod position prediction,and is applicable to various advanced micro reactors at various operating conditions.