加密貨幣具有去中心化、交易紀錄公開透明無法任意竄改，且發行總量固定的特性，從最早的加密貨幣——比特幣問世以來，市場對於加密貨幣的關注熱度日益興起，帶來了許多新的交易模式與投資型態。然而，加密貨幣的價格起伏震盪劇烈，對於投資人而言，若能事先預測價格走勢以規劃資金的投入與分配，將可以避免突如其來的鉅額賠損，確保財產的安全。
本研究以比特幣為研究對象，探討結合訊號分解方法與深度學習模型預測幣特幣價格的可能性，使用Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN)、Variational Mode Decomposition (VMD)、Ensemble Patch Transform (EPT) 三種訊號分解方法，以及EPT-CEEMDAN、EPT-VMD兩種複合式訊號分解方法，將2019到2022年比特幣的小時價格原始資料分解為數個子分量，再個別輸入到深度學習模型Long Short Term Memory (LSTM)、Bidirectional Long Short Term Memory (BiLSTM)與Temporal Convolutional Network (TCN)中進行訓練與預測，最後將所有子分量的預測值重建回最終預測價格，並比較各種實驗組合的預測誤差率。經實驗證實在多數組合情況下，結合訊號分解方法可以有效提升模型的預測性能。此外，通過實驗找出在四年的資料中，綜合預測性能最佳的實驗組合為EPT-VMD-TCN方法，四年之預測MAPE誤差率分別為0.0014、0.004、0.0007和0.0076。

關鍵詞：比特幣、訊號分解、深度學習、時間序列預測

Cryptocurrency is a new form of currency that is decentralized, with transaction records that are completely open and cannot be tampered. Since the emergence of the first cryptocurrency, Bitcoin, the market's interest in cryptocurrencies has been steadily increasing. This has opened up many new investment opportunities and trading methods. However, cryptocurrency prices tend to be highly volatile. For investors, being able to predict the price trends of cryptocurrencies allows for advance planning of fund allocation and investment, avoiding sudden and substantial losses, and ensuring the safety of assets.
In this study, focus on Bitcoin as the research subject and investigates the feasibility of combining signal decomposition methods with deep learning models to predict the price of Bitcoin. Using signal decomposition methods like Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational Mode Decomposition (VMD), Ensemble Patch Transform (EPT) and hybrid methods: EPT-CEEMDAN, EPT-VMD to decompose the hourly price data of Bitcoin from 2019 to 2022 into several sub-components. These sub-components were then separately inputted into deep learning models, Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM) and Temporal Convolutional Network (TCN), for training and prediction. Finally, the predicted values of all sub-components were reconstructed to obtain the final predicted price. The prediction errors of various experimental combinations were compared.
The results showed that combining signal decomposition methods with deep learning models can effectively enhance the predictive performance of the models. Additionally, in the experiments, it was found that the best-performing method among four years of data was the EPT-VMD-TCN method, with MAPE (Mean Absolute Percentage Error) rates of 0.0014, 0.004, 0.0007, and 0.0076 for the four years, respectively.

Keywords: Bitcoin, Signal Decomposition, Deep Learning, Time Series Forecasting

[1] 蕭琮峻（2022）。應用人工智慧方法於加密貨幣價格預測之研究。﹝碩士論文。國立臺北大學﹞臺灣博碩士論文知識加值系統。 https://hdl.handle.net/11296/m3fn42
[2] Aina, L., Gulordava, K., & Boleda, G. (2019). Putting words in context: LSTM language models and lexical ambiguity. arXiv preprint arXiv:1906.05149.
[3] Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.
[4] Carvalho, V. R., Moraes, M. F., Braga, A. P., & Mendes, E. M. (2020). Evaluating five different adaptive decomposition methods for EEG signal seizure detection and classification. Biomedical Signal Processing and Control, 62, 102073.
[5] Cao, J., Li, Z., & Li, J. (2019). Financial time series forecasting model based on CEEMDAN and LSTM. Physica A: Statistical mechanics and its applications, 519, 127-139.
[6] Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014). Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555.
[7] Cid-Sueiro, J., Arribas, J. I., Urbán-Munoz, S., & Figueiras-Vidal, A. R. (1999). Cost functions to estimate a posteriori probabilities in multiclass problems. IEEE Transactions on Neural Networks, 10(3), 645-656
[8] Dragomiretskiy, K., & Zosso, D. (2013). Variational mode decomposition. IEEE transactions on signal processing, 62(3), 531-544.
[9] Du, X., Tang, Z., Wu, J., Chen, K., & Cai, Y. (2022). A New Hybrid Cryptocurrency Returns Forecasting Method Based on Multiscale Decomposition and an Optimized Extreme Learning Machine Using the Sparrow Search Algorithm. IEEE Access, 10, 60397-60411.
[10] Ertam, F. (2019). An effective gender recognition approach using voice data via deeper LSTM networks. Applied Acoustics, 156, 351-358.
[11] Giles, C. L., Kuhn, G. M., & Williams, R. J. (1994). Dynamic recurrent neural networks: Theory and applications. IEEE Transactions on Neural Networks, 5(2), 153-156.
[12] Glorot, X., Bordes, A., & Bengio, Y. (2011, June). Deep sparse rectifier neural networks. In Proceedings of the fourteenth international conference on artificial intelligence and statistics (pp. 315-323). JMLR Workshop and Conference Proceedings.
[13] Guo, C., Kang, X., Xiong, J., & Wu, J. (2022). A New Time Series Forecasting Model Based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Temporal Convolutional Network. Neural Processing Letters, 1-21.
[14] Gyamerah, S. A. (2022). On forecasting the intraday Bitcoin price using ensemble of variational mode decomposition and generalized additive model. Journal of King Saud University-Computer and Information Sciences, 34(3), 1003-1009.
[15] Haykin, S. (1998). Neural networks: a comprehensive foundation. Prentice Hall PTR.
[16] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[17] Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H. H., Zheng, Q., ... & Liu, H. H. (1998). The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proceedings of the Royal Society of London. Series A: mathematical, physical and engineering sciences, 454(1971), 903-995.
[18] Huang, N. E., Wu, M. L., Qu, W., Long, S. R., & Shen, S. S. (2003). Applications of Hilbert–Huang transform to non‐stationary financial time series analysis. Applied stochastic models in business and industry, 19(3), 245-268.
[19] Kavinnilaa, J., Hemalatha, E., Jacob, M. S., & Dhanalakshmi, R. (2021, July). Stock price prediction based on LSTM deep learning model. In 2021 International Conference on System, Computation, Automation and Networking (ICSCAN) (pp. 1-4). IEEE.
[20] Kim, D., Choi, G., & Oh, H. S. (2020). Ensemble patch transformation: a flexible framework for decomposition and filtering of signal. EURASIP Journal on Advances in Signal Processing, 2020, 1-27.
[21] Kim, D., Oh, H. S., & Choi, G. (2021). EPT: An R package for ensemble patch transform. SoftwareX, 14, 100704.
[22] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[23] Li, D., Jiang, F., Chen, M., & Qian, T. (2022). Multi-step-ahead wind speed forecasting based on a hybrid decomposition method and temporal convolutional networks. Energy, 238, 121981.
[24] Lindeberg, T. (2013). Scale-space theory in computer vision (Vol. 256). Springer Science & Business Media.
[25] Lutfi, M., Agustin, S. P., & Yulita, I. N. (2021, October). LQ45 Stock Price Prediction Using Linear Regression Algorithm, Smo Regression, And Random Forest. In 2021 International Conference on Artificial Intelligence and Big Data Analytics (pp. 1-5). IEEE.
[26] Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Decentralized business review, 21260.
[27] Nielsen, M. A. (2015). Neural networks and deep learning (Vol. 25, pp. 15-24). San Francisco, CA, USA: Determination press.
[28] Nina B., Steven E.(2023) Blockcast.it . Retrieved from https://blockcast.it/2023/01/10/ftx-reached-out-to-tether-for-economic-help-before-collapse/
[29] Olah, C. (2015). Understanding LSTM Networks. Retrieved from http://colah.github.io/posts/2015-08-Understanding-LSTMs/
[30] Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE transactions on Signal Processing, 45(11), 2673-2681.
[31] Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. (2014). Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15(1), 1929-1958.
[32] Torres, M. E., Colominas, M. A., Schlotthauer, G., & Flandrin, P. (2011, May). A complete ensemble empirical mode decomposition with adaptive noise. In 2011 IEEE international conference on acoustics, speech and signal processing (ICASSP) (pp. 4144-4147). IEEE.
[33] Wang, M. (2020). Bitcoin and its impact on the economy. arXiv preprint arXiv:2010.01337.
[34] Wijesinghe, G. W. R. I., & Rathnayaka, R. M. K. T. (2020, December). ARIMA and ANN Approach for forecasting daily stock price fluctuations of industries in Colombo Stock Exchange, Sri Lanka. In 2020 5th International Conference on Information Technology Research (ICITR) (pp. 1-7). IEEE.
[35] Wu, Z., & Huang, N. E. (2009). Ensemble empirical mode decomposition: a noise-assisted data analysis method. Advances in adaptive data analysis, 1(01), 1-41.
[36] Zan, S., & Zhang, Q. (2023). Short-Term Power Load Forecasting Based on an EPT-VMD-TCN-TPA Model. Applied Sciences, 13(7), 4462.