臺灣博碩士論文加值系統

由於股票市場受到許多複雜和不確定因素的影響，股票價格預測一直是具有挑戰性的任務。傳統的股票價格預測方法通常基於技術分析或基本面分析，但這些方法通常很依賴專家的知識和經驗。近年來，深度學習應用於股票價格預測呈現指數性的成長。CNN的卷積運算能夠提取數據特徵，而池化操作透過減少數據特徵的尺寸和數量，不但保留最重要的特徵，也同時減少計算量，提高模型的效率和泛化能力。LSTM透過記憶單元和閘控機制，在具有長期或短期依賴性序列數據的預測上皆表現良好。利用以上兩個模型的優勢，先使用CNN來提取序列數據的特徵，再使用LSTM來學習特徵序列的長期或短期依賴性，使得CNN-LSTM在預測股票價格等序列數據方面優於LSTM。
為了提高CNN-LSTM[14]預測的準確性，將VGG11原本的卷積參數進行調整並刪減兩層的池化層，命名為MVGG11（Modified Visual Geometry Group 11）。這樣的改變旨在減少VGG的計算量，同時提高CNN-LSTM的CNN特徵提取能力。由於Inception-v2模組在卷積層中使用多個不同大小的卷積核，並且以並行的方式進行多個卷積運算，從而讓模型同時提取不同大小的特徵，增強了模型的預測能力。因此，本研究更進一步將Inception-v2模組A、B和C分別引入MVGG11並結合LSTM提出MVGG11-MA-LSTM、MVGG11-MB-LSTM和MVGG11-MC-LSTM三個模型。
實驗結果顯示CNN-LSTM[14]的R^2為0.877661，其他各項評估指標也都優於RNN、GRU和LSTM，CNN-LSTM的R^2值比LSTM提升8.8%。而MVGG11-LSTM比CNN-LSTM的R^2值降低3.31%。然而引入Inception-v2模組A、B和C後R^2值比CNN-LSTM提高1.6%到3.28%，其中又以MVGG11-MB-LSTM的各項評估指標最高，R^2達到0.906436。

Since the stock market is affected by many complex and uncertain factors, stock price forecasting has always been a challenging task. Traditional stock price prediction methods are usually based on technical analysis or fundamental analysis, but these methods usually rely heavily on the knowledge and experience of experts. In recent years, the application of deep learning to stock price prediction has shown exponential growth. The convolution operation of the CNN can extract data features, and the pooling operation not only retains the most important features by reducing the size and number of data features, but also reduces the amount of calculation, improving the efficiency and generalization ability of the model. The LSTM performs well in the prediction of sequence data with long-term or short-term dependencies through memory cells and gating mechanisms. Taking advantage of the above two models, first use CNN to extract the features of sequence data, and then use LSTM to learn the long-term or short-term dependence of feature sequences, making CNN-LSTM better than LSTM in predicting sequence data such as stock prices.
In order to improve the prediction accuracy of CNN-LSTM [14], the original convolution parameters of VGG11 were adjusted and the two-layer pooling layer was deleted, named MVGG11 (Modified Visual Geometry Group 11). Such changes aim to reduce the computational load of VGG while improving the CNN feature extraction capability of CNN-LSTM. Since the Inception-v2 module uses multiple convolution kernels of different sizes in the convolution layer, and performs multiple convolution operations in parallel, the model can extract features of different sizes at the same time, which enhances the predictive ability of the model. Therefore, this study further introduces Inception-v2 modules A, B and C into MVGG11 respectively, and combines LSTM to propose three models, MVGG11-MA-LSTM, MVGG11-MB-LSTM and MVGG11-MC-LSTM.
The experimental results show that the R^2 of CNN-LSTM [14] is 0.877661, and other evaluation indicators are also better than RNN, GRU and LSTM. The R^2 value of CNN-LSTM is 8.8% higher than that of LSTM. The R^2 value of MVGG11-LSTM is 3.31% lower than that of CNN-LSTM. However, after introducing Inception-v2 modules A, B, and C, the R^2 value is 1.6% to 3.28% higher than that of CNN-LSTM. Among them, MVGG11-MB-LSTM has the highest evaluation index, and R^2 reaches 0.906436.

摘要.............................................................................................................................................i
Abstract.......................................................................................................................................ii
誌謝...........................................................................................................................................iv
目錄............................................................................................................................................v
表目錄 ......................................................................................................................................vii
圖目錄 .....................................................................................................................................viii
第一章 緒論..............................................................................................................................1
第二章 相關深度學習模型介紹 ..............................................................................................2
2.1 循環神經網路（Recurrent Neural Network，RNN） ...................................................2
2.2 長短期記憶 (Long Short-Term Memory，LSTM) ........................................................2
2.3 閘控循環單元 (Gated Recurrent Unit，GRU)................................................................4
2.4 卷積神經網路 (Convolutional Neural Network，CNN)................................................5
2.5 LeNet.................................................................................................................................6
2.6 AlexNet.............................................................................................................................6
2.7 VGG (Visual Geometry Group)........................................................................................7
2.8 Inception-v1 ......................................................................................................................7
2.9 Inception-v2 模組 A .........................................................................................................8
2.10 Inception-v2 模組 B .......................................................................................................9
2.11 Inception-v2 模組 C .....................................................................................................10
2.12 一維卷積神經網路(One-Dimensional Convolutional Neural Network，1D-CNN)..11
2.13 CNN-LSTM..................................................................................................................11
第三章 一維卷積神經網路優化結合長短期記憶模型 ........................................................13
3.1 MVGG11-LSTM ............................................................................................................13
3.2 MVGG11-MA-LSTM.....................................................................................................14
3.3 MVGG11-MB-LSTM.....................................................................................................15
3.4 MVGG11-MC-LSTM.....................................................................................................17
3.5 模型參數設計................................................................................................................18
第四章 實驗結果與討論 ........................................................................................................29
4.1 開發環境........................................................................................................................29
4.2 實驗步驟.........................................................................................................................29
4.3 數據集與評估指標........................................................................................................29
4.4 實驗結果對比分析........................................................................................................31
第五章 結論..........................................................................................................................49
參考文獻 ..................................................................................................................................50
Extended Abstract.....................................................................................................................52
Abstract.....................................................................................................................................52
1. Introduction .......................................................................................................................54
2. Experimental Method........................................................................................................54
3. Experimental Results.........................................................................................................54
4. Conclusion.........................................................................................................................55
Reference ..................................................................................................................................56

[1] R Vinayakumar; E. A Gopalakrishnan; Vijay Krishna Menon; K. P. Soman, 2017, ” Stock price prediction using LSTM, RNN and CNN-sliding window model”, 2017 international conference on advances in computing communications and informatics (icacci), pp. 1643-1647.
[2] Kavinnilaa J; Hemalatha E; Minu Susan Jacob; Dhanalakshmi R, 2021, ”Stock Price Prediction Based on LSTM Deep Learning Model”, 2021 International Conference on System, Computation, Automation and Networking (ICSCAN), pp. 451-455.
[3] Pineda, F. J., 1987, Generalization of back-propagation to recurrent neural networks. Physical Review Letters, 19(59):2229-2232.
[4] S. Hochreiter, Y. Bengio, P. Frasconi, and J. Schmidhuber., 2001, ”Gradient Flow in Recurrent Nets: the Difficulty of Learning Long-term Dependencies.”, In S. C. Kremer and J. F. Kolen, editors, A Field Guide to Dynamical Recurrent Neural Networks.
[5] Hochreiter & Schmidhuber, 1997, ” Long Short-Term Memory”, Neural Computation，Volume: 9, Issue: 8, Page(s): 1735 – 1780.
[6] Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, Yoshua Bengio, 2014, ”Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling”, arXiv:1412.3555.
[7] Hao Y, Gao Q , 2020, ”Predicting the trend of stock market index using the hybrid neural network based on multiple time scale feature learning”, Appl Sci 10(11):3961–3974.
[8] Y. Lecun; L. Bottou; Y. Bengio; P. Haffner, 1998, ” Gradient-based learning applied to document recognition”, Proceedings of the IEEE , Volume: 86, Issue: 11.
[9] Karen Simonyan & Andrew Zisserman, 2014, ” Very Deep Convolutional Networks for Large-Scale Image Recognition”, arXiv:1409.1556.
[10] Krizhevsky, A., Sutskever, I., & Hinton, G. E., 2012, ”ImageNet classification with deep convolutional neural networks”, In Advances in neural information processing systems, pp. 1097-1105.
[11] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna, 2015, ”Rethinking the Inception Architecture for Computer Vision”, In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818-2826.
[12] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich, 2014, ”Going Deeper with Convolutions”, arXiv:1409.4842.
[13] Gülyeter ÖZTÜRK; Raşit KÖKER; Osman ELDOğAN; Durmuş KARAYEL, 2020, ”Recognition of Vehicles, Pedestrians and Traffic Signs Using Convolutional Neural Networks”, 2020 4th International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), pp. 1-8, doi: 10.1109/ISMSIT50672.2020.9255148.
[14] Lu, W., Li, J., Li, Y., Sun, A. & Wang, J., 2020, ” A CNN-LSTM-Based Model to Forecast Stock Prices”, Complexity (New York, N.Y.), vol. 2020.
[15] Y. Yang et al., 2021, ” Implementation of Reconfigurable CNN-LSTM Accelerator Based on FPGA”, 2021 IEEE 21st International Conference on Communication Technology (ICCT), pp. 1026-1030, doi: 10.1109/ICCT52962.2021.9657920.
[16] Ta V, Liu C, Tadesse D , 2020, ”Portfolio optimization-based stock prediction using long-short term memory network in quantitative trading”, Appl Sci 10(2):437–456.
[17] J. Wang, Y. Wang and J. Yang, 2021, ”Forecasting of significant wave height based on gated recurrent unit network in the taiwan strait and its adjacent waters”, Water, vol. 13, no. 1.
[18] Lavin J. Halawa; Adi Wibowo; Ferda Ernawan, 2019, ”Face Recognition Using Faster R-CNN with Inception-V2 Architecture for CCTV Camera”, 3rd International Conference on Informatics and Computational Sciences (ICICoS) , pp. 1-6.
[19] https://cn.investing.com/equities/intertrust-group-holding-sa-historical-data