罗马5期耀变体多波段目录BZUs的分类  

Classication for BZUs in 5BZCAT

作　　者：朱惊天 樊军辉[1,2,3] 蔡金庭 Zhu Jingtian;Fan Junhui;Cai Jinting(Center for Astrophysics,Guangzhou University,Guangzhou 510006,China;Astronomy Science and Technology Research Laboratory of Department of Education of Guangdong Province,Guangzhou 510006,China;Key Laboratory for Astronomical Observation and Technology of Guangzhou,Guangzhou 510006,China)

机构地区：[1]广州大学天体物理中心,广东广州510006 [2]广东省高校天文观测与技术重点实验室,广东广州510006 [3]广州天文观测与技术重点实验室,广东广州510006

出　　处：《天文研究与技术》2021年第1期25-42,共18页Astronomical Research & Technology

基　　金：国家自然科学基金(11733001,U1531245);广东省自然科学基金(2017A030313011,2019B030302001);广东省和广州市重点学科资助.

摘　　要：利用支持向量机(Support Vector Machine,SVM)、随机森林(Random Forest,RF)、集成学习(Ensemble Learning,EM)和多层感知机(Multi-Layer Perceptron,MLP)将罗马5期耀变体多波段目录(The 5th edition of the Roma-BZCAT Multifrequency Catalogue of Blazars,5BZCAT)中227个不确定类型的耀变体(Blazars of Uncertain type,BZUs)分为蝎虎天体候选体和平谱射电类星体候选体,并通过特征工程和网格搜索方法提高分类准确率。综合4种分类器的分类结果,将判别概率阈值设为0.8,得到33个蝎虎天体候选体和119个平谱射电类星体候选体。In order to evaluate the potential optical classification of 227 BZUs in 5BZCAT,we divide the BZUs into BL lac candidates and FSRQ candidates by four machine learning methods:support vector machine(SVM),random forest(RF),ensemble learning(EM)and multi-layer perceptron(MLP).And the classification accuracy is improved by feature engineering and grid search.By combining the classification results of four classifiers and setting the threshold of discrimination probability to 0.8,we get 33 BL lacs candidates and 119 FSRQs candidates.

关 键 词：耀变体 蝎虎天体 平谱射电类星体 机器学习 分类 

分 类 号：TN216[电子电信—物理电子学]