摘 要：脉冲星是一种具有极大的科学价值的中子星。其在旋转时，地球上可探测到其发射的无线电，而且这种辐射存在周期性。对脉冲星的分类主要采用机器学习算法，各种算法表现不同。该研究对K 邻近，决策树，朴素贝叶斯，梯度提升树算法对脉冲星信号进行了二分类，基于F1 值和AUC 值评估模型，算法对脉冲星的旋转轨迹进行研究，以此来判定各种分类算法的表现，结果显示对于未调参的算法，逻辑回归表现最好，梯度提升树其次。K 邻近和决策树表现相对较差。
关键词：脉冲星；脉冲星候选样本；机器学习；分类算法；二分类
中图分类号：TP181；TP391.41 文献标识码：A 文章编号：2096-4706（2020）07-0076-04
Pulsar Detection Based on Machine Learning Algorithm
ZHOU Yu
（Xuzhou University of Technology，Xuzhou 221018，China）
Abstract：Pulsar is a neutron star with great scientific value. When it rotates，the radio it sends can be detected on the earth，and the radiation is periodic. The classification of pulsars mainly uses machine learning algorithm，and each algorithm has different performance. In this study，K-neighborhood，decision tree，naive Bayes，and gradient lifting tree algorithm are used to classify pulsar signals. Based on the evaluation model of F1 and AUC values，the algorithm studies the rotation path of pulsar，so as to determine the performance of various classification algorithms. The results show that for the algorithm without parameter adjustment，logical regression is the best，and gradient lifting tree is the second. K-proximity and decision tree performance are relatively poor.
Keywords：pulsar；pulsar candidate samples；machine learning；classification algorithm；binary classification
参考文献：
[1] BATES S D，BAILES M，BARSDELL B R，et al. The High Time Resolution Universe Pulsar Survey- VI. An artificial neural network and timing of 75 pulsars [J].Monthly Notices of the Royal Astronomical Society，2012，427（2）：1052-1065.
[2] LEVIN L，BAILES M，BATES S D，et al. Radio emission evolution，polarimetry and multifrequency single pulse analysis of the radio magnetar PSR J1622?4950 [J].Monthly Notices of the Royal
Astronomical Society，2012，422（3）：2489-2500.
[3] KEITH M J，JOHNSTON S，BAILES M，et al. The High Time Resolution Universe Pulsar Survey-IV. Discovery and polarimetry
of millisecond pulsars [J].Monthly Notices of the Royal Astronomical Society，2012，419（2）：1752-1765.
[4] 王东刚. 残余引力波：引力子、真空与规则化 [D]. 合肥：中国科学技术大学，2016.
[5] 刘良端. 双中子星并合产生的电磁信号研究 [D]. 南京：南京大学，2016.
[6] 张璐. 高频引力波电磁谐振效应的最优参数选择与噪声初步分析 [D]. 重庆：重庆大学，2016.
作者简介：周宇（1979.06—），男，汉族，江苏徐州人，工程师，硕士，研究方向：计算机。