為了搜尋雙星系統，我們使用JKTEBOP程式[10]建構資料庫。然後構建了三種不同干擾系數的模型，並訓練它們在卷積神經網路的運算中。光曲線數據做為模型的輸入。我們利用在卷積神經網路中不同的參數建構一個最佳模型，利用此模型在三種不同的干擾背景之下，對它們的數值結果進行了比較和改進。最後結果，我們發現望遠鏡的干擾背景對雙星系統的參數模型有著重要影響。

In order to search the binary star system, we use the JKTEBOP program [10] to construct the database. Then we construct three models with different noises coefficients and train them in the operation of convolutional neural network. The light curve data is used as the input of the model. We use different parameters in convolutional neural network to construct an optimal model, and use this model to compare and improve their numerical results under three different noises. As a result, we found that the interference background of the telescope has an important influence on the parameter model of the binary system.

致謝 II
摘要 III
Abstract IV
Chapter 1 Introduction 2
Chapter 2 Machine Learning 4
2.1 Neural Networks Architecture (NN) 4
2.1.1 The Loss Functions 6
2.1.2 The Activation Functions 7
2.2 Convolutional Neural Network(CNN) 10
2.2.1 Pooling Layer 12
2.2.2 Keras 13
2.3 Eclipsing Binaries Model 15
Chapter 3 Models and Parameters 23
3.1 Activation Functions and Optimizer Methods 23
3.1.1 Two-layers 24
3.1.2 Three-layers 25
3.1.3 Four-layers 26
3.2 Total Data Size 27
Chapter 4 Results 30
4.1 Precision, Recall, and F score 30
4.2 Models 32
4.2.1 Model A 33
4.2.2 Model B 35
4.2.3 Model C 38
4.3 The Comparison and Improvement of Three Models 41
Chapter 5 Conclusions 44
References 45
Appendix A Optimizer 47
Appendix B Program 52

[1] Aitken, Robert G, The Binary Stars, New York: McGraw-Hill, 1935:1.
[2] Robert Grant Aitken, The Binary Stars, New York: Dover, 1964, p.1.
[3] JKTEBOP, The solar-type eclipsing binary system LL Aquarii, Astronomy & Astrophysics, September 2013, Volume 557.
[4] https://mropengate.blogspot.com/2015/06/ch15-4-neural-network.html
[5] https://medium.com/%E5%AD%B8%E4%BB%A5%E5%BB%A3%E6%89%8D/%E5%84%AA%E5%8C%96%E6%BC%94%E7%AE%97-3-gradient-descent-with-momentum-52a65550edc0
[6] Chintarungruangchai, Pattana and Jiang, Guey, Detecting Exoplanet
Transits through Machine-learning Techniques with Convolutional Neural Networks, Publications of the Astronomical Society of the Pacific, 2019, vol 131, 064502.
[7] Prs ̌a, A and Guinan, EF and Devinney, EJ and DeGeorge, M and Bradstreet, DH and Giammarco, JM and Alcock, CR and Engle, SG, Artificial intelligence approach to the determination of physical properties of eclipsing binaries. i. the ebai project, The Astrophysical Journal, 2008, vol 687, 542.
[8] Hinners, Trisha A and Tat, Kevin and Thorp, Rachel, Machine learning techniques for stellar light curve classification, The Astronomical Journal, 2018, vol 156, 7.
[9] Ketkar, Nikhil and others, Deep Learning with Python, Springer, 2017.
[10] Southworth, John, JKTEBOP: Analyzing light curves of detached eclipsing binaries, Astrophysics Source Code Library, 2012.