基于BP神经网络的核临界安全数据实时计算方法

Real-Time Calculation of Nuclear Critical Safety Data Based on BP Neural Network

基于核燃料后处理工艺过程中的多体系统,建立高精度BP神经网络预测模型,可实现核临界安全参数keff的实时智能预测,辅助对工艺系统的运行状态及安全性能的进行快速评估。该方法利用MCNP模拟计算得到的多体系统的核临界安全原始数据集,通过机器学习(ML)的方法实现复杂体系的建模,能够克服多系统后处理工艺设备运行参数联动变化带来的实时定量评估困难。数值实验表明,预测系统的误差在1%以内,计算速度比MCNP模拟快2000倍以上,是ML等智能方法在核临界安全领域重要的探索,具有广阔的前景。

Based on the multi-body system in the nuclear fuel reprocessing, a high-precision BP neural network prediction model is established. It can realize the real-time intelligent prediction of the nuclear critical safety parameter keff, and assist the rapid auxiliary evaluation of the operation status and safety performance of the process system. This method uses the original data set of nuclear criticality safety obtained by MCNP simulation, and realizes the modeling of complex system by machine learning (ML), which can overcome the difficulty of real-time quantitative evaluation caused by the linkage change of operation parameters of multi-system post-processing equipment. Numerical experiments show that the error of the prediction system is less than 1%, and the calculation speed is more than 2000 times faster than that of MCNP simulation. It is an important exploration of ML and other intelligent methods in the field of nuclear critical safety, and has broad prospects.