基于机器学习的燃料棒核素及燃耗预测

Prediction of fuel rod nuclide and burnup based on machine learning

为提升核反应堆燃料棒的燃耗预测能力,采用机器学习方法,依据数值计算VVER-1000型反应堆得到的U41型燃料棒的核素含量及平均燃耗数据样本,通过岭回归、BP神经网络和卷积神经网络3种算法,建立不同平均燃耗与核素含量间的回归模型,并以均方误差(MSE)及R 2作为评估标准评价模型,利用训练好的模型在测试集中对目标进行预测。结果表明:岭回归、BP神经网络及卷积神经网络等机器学习方法在预测核素含量及平均燃耗方面有较高准确性,与传统方法相比,降低了整个测量过程的复杂程度,提高了测量效率,可为人工智能算法在核工业领域的应用提供参考。

In order to improve the accuracy of predicting the burnup of the fuel rod,the machine learning was used.According to the nuclide content and average burnup data samples of U41 fuel rod obtained by numerical simulation of VVER-1000 reactor,the three algorithms of ridge regression,BP neural network and convolutional neural network were used to establish the regression model between different average burnup and nuclide content.Mean Squared Error(MSE)and R 2 were used as evaluation standard evaluation models.The trained models were used to predict the target in the test set,and the results show that machine learning methods such as ridge regression,BP neural network and convolutional neural network have high accuracy in predicting nuclide content and average burnup,which provides an application of artificial intelligence algorithms in the nuclear industry.At the same time,compared with traditional methods,the complexity of the entire measurement process was reduced,and the measurement efficiency was improved.