论文标题
部分可观测时空混沌系统的无模型预测
Stellar mass and radius estimation using Artificial Intelligence
论文作者
论文摘要
对于大多数恒星而言，估计恒星质量和半径是一个挑战，但是它们的知识对于许多不同的天体物理领域至关重要。估计这些变量的最扩展的技术之一是所谓的经验关系。在这项工作中，我们提出了一组最先进的AI回归模型，目的是研究它们在估计恒星质量和半径方面的熟练程度。我们公开发布数据库，AI模型以及对社区恒星质量和半径估算的在线工具。我们在文献中使用726毫秒恒星的样本，具有准确的M，R，T_EFF，L，LOG G和[Fe/H]。我们已经将数据样本分为培训和测试集，然后与它们分析了不同的AI技术。特别是，我们已经通过实验评估了以下模型的准确性：线性reg。，贝叶斯注册，回归树，随机森林，支撑量为reg。 （SVR），神经网络，KNN和堆叠。我们提出了一系列旨在评估估计准确性的实验。我们还分析了减少输入参数数量的影响，并将我们的结果与文献中最先进的经验关系的结果进行了比较。我们发现，堆叠几种回归模型是估计质量和半径的最合适技术。对于质量，神经网络也提供精确的结果，对于半径，SVR和神经网络也可以工作。在与其他基于经验关系的模型进行比较时，我们的堆叠量将两个变量的精度提高了两个。另外，在恒星质量的情况下，偏差将减少为一个数量级。最后，我们发现，使用我们的堆叠，仅使用T_EFF和L作为输入特征，所获得的精度略大于5％，偏置约1.5％。
Estimating stellar masses and radii is a challenge for most of the stars but their knowledge is critical for many different astrophysical fields. One of the most extended techniques for estimating these variables are the so-called empirical relations. In this work we propose a group of state-of-the-art AI regression models, with the aim of studying their proficiency in estimating stellar masses and radii. We publicly release the database, the AI models, and an online tool for stellar mass and radius estimation to the community. We use a sample of 726 MS stars in the literature with accurate M, R, T_eff, L, log g, and [Fe/H]. We have split our data sample into training and testing sets and then analyzed the different AI techniques with them. In particular, we have experimentally evaluated the accuracy of the following models: Linear Reg., Bayesian Reg., Regression Trees, Random Forest, Support-Vector Reg. (SVR), Neural Networks, kNN, and Stacking. We propose a series of experiments designed to evaluate the accuracy of the estimations. We have also analyzed the impact of reducing the number of inputs parameters and compared our results with those from state-of-the-art empirical relations in the literature. We have found that a Stacking of several regression models is the most suitable technique for estimating masses and radii. In the case of the mass, Neural Networks also provide precise results, and for the radius, SVR and Neural Networks work too. When comparing with other state-of-the-art empirical relations based models, our Stacking improves the accuracy by a factor of two for both variables. In addition, bias is reduced to one order of magnitude in the case of the stellar mass. Finally, we have found that using our Stacking and only T_eff and L as input features, the accuracies obtained are slightly larger than a 5%, with a bias approx 1.5%.
