利用高斯过程回归对燃爆单元宽度的预测方法研究  

Methodological Study on Prediction of Detonation Cell Width with Gaussian Process Regression

作　　者：侯炳旭[1] 俞冀阳[1] 徐沾杰 江光明[3] 邹志强[3] 

机构地区：[1]清华大学工程物理系,北京100084 [2]卡尔斯鲁厄理工学院 [3]中国核动力研究设计院核反应堆系统设计技术重点实验室,成都610213

出　　处：《核动力工程》2017年第2期72-77,共6页Nuclear Power Engineering

基　　金：核反应堆系统设计技术重点实验室项目运行基金(HT-A100K-02-201406)

摘　　要：燃爆单元宽度(λ)是度量可燃气体燃爆风险的一项重要参数。文中把λ和特征化学反应区厚度(δ)联系起来,以无量纲活化能和无量纲温度为自变量,以λ/δ的对数为因变量对实验数据进行回归。针对传统参数回归方法的不足,采用基于机器学习的高斯过程回归(GPR)方法完成数据拟合工作。通过比较实验数据和拟合函数的预测值,发现GPR方法的结果能够较为准确地预测不同组分的可燃混合气体在不同初始条件下气体的λ。与传统参数回归的结果相比,GPR方法在拟合精度上优于传统参数回归方法。Detonation cell width is an important parameter to quantify the detonation risk of a flammable gas mixture. In this study, detonation cell width, λ, is correlated with the characteristic chemical reaction zone width, δ. Regression is performed by using experimental data, with non-dimensional activation energy and non-dimensional temperature as independent variables and the logarithm of λ/δ as a dependent variable. Considering the disadvantages of traditional parametric regression methods, a Gaussian process regression based on machine learning, is applied in this study to fit the experimental data. By the comparison of the measured data and predictions of the regression result, it is found that the Gaussian process regression method can precisely predict the detonation cell width of flammable gas mixtures with different compositions and at different initial conditions. Meanwhile, the Gaussian process regression method takes an advantage in fitting precision comparing to the traditional parametric regression methods.

关 键 词：氢气燃爆 燃爆单元宽度(λ) 高斯过程回归(GPR) 

分 类 号：TL364[核科学技术—核技术及应用]