學術產出-Theses
Article View/Open
Publication Export
-
題名 時間序列資料的相關性異常偵測
Anomaly Detection of Correlation Change over Multivariate Time Series作者 曹昱維
Tsao, Yu-Wei貢獻者 沈錳坤
Shan, Man-Kwan
曹昱維
Tsao, Yu-Wei關鍵詞 多變輛時間序列
相關性預測
相關性異常偵測
Multivariate Time Series
Correlation Prediction
Correlation Anomaly Detection日期 2024 上傳時間 1-Mar-2024 13:42:40 (UTC+8) 摘要 多變量時間序列的研究包括預測、分類、分群和異常偵測等。其中，時間序列異常偵測是近年熱門的研究主題。其目標在早期發現時間序列中的異常現象。針對多變量時間序列的相關性異常偵測，很少現有研究。本研究的主要目的是偵測多變量時間序列資料的相關性異常。 本研究採取非監督式學習，根據正常多變量序列所學習的相關性預測模型，來偵測相關性異常。本研究提出 CARG模型 (Clustering-Attention-Residual-GRU Model)來預測多變量時間序列之間的相關性。CARG 模型不僅能夠捕捉相關性本身的時間性結構，同時還透過學習多變量之間的相依性資訊，從而提高預測效果。CARG 模型結合了時間序列分群、注意力機制和時間序列分析模型等技術，具有能夠有效捕捉多變量之間相依性結構的優勢。當 CARG 模型的預測值與實際值存在顯著差異時，可能發生相關性異常事件。 在我們的實驗中，我們將 CARG 模型應用於金融市場資料上，CARG 模型展現出相當不錯的表現。此外，實驗結果也表明，在 CARG 模型中所設計的各結構對多變量時間序列相關性預測任務均有所助益。並且透過與Baseline模型的比較，CARG 模型確實加強預測的效能。
The research domain of multivariate time series consists of forecasting, classification, clustering, and anomaly detection, among others. Anomaly detection in time series has become a popular research topic in recent years. Its goal is to discover anomalies in time series data at an early stage. Little research has been paid on the anomaly detection of correlations in multivariate time series. The primary objective of this thesis is to detect correlation anomalies in multivariate time series data. This research adopts unsupervised learning approach to detect correlation anomalies based on a correlation prediction model learned from normal multivariate series. We propose the CARG model (Clustering-Attention-Residual-GRU model) to predict the correlations among multivariate time series. The CARG model not only captures the temporal structure of the correlations but also improves prediction performance by learning the dependency information among the multivariates. Combining techniques of time series clustering, attention mechanisms, and time series analysis models, the CARG model is advantageous in effectively capturing the dependency structure among multivariates. To detect correlation anomalies, a significant discrepancy between the predicted values by the CARG model and the actual values may indicate a correlation anomaly event. In our experiments, we applied the CARG model to financial market data and it showed quite impressive performance. Additionally, the experimental results also demonstrate that the various structures designed in the CARG model contribute positively to the task of predicting correlations in multivariate time series. By comparison with baseline models, the CARG model indeed enhances the effectiveness of prediction.參考文獻 [1] A. Blázquez-García, A. Conde, U. Mori, and J. A. Lozano, A review on outlier/anomaly detection in time series data, ACM Computing Surveys, Vol. 54, No. 3, 2021. [2] A. Deng, and B. Hooi, Graph neural network-based anomaly detection in multivariate time series, Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35, No. 5, 2021. [3] A. Hanni, Correlation-based anomaly detection in time series, Master Thesis, Department of Informatics, Universiy of Bern, 2020. [4] A. L. Goldberger, et al., Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals, Circulation, Vol. 101, No. 23, 2000. [5] A. Vaswani, et al., Attention is all you need, Advances in Neural Information Processing Systems (NIPS), Vol. 30, 2017. [6] C. Zhang, S. Bengio, M. Hardt, B. Recht, and O. Vinyals, Understanding deep learning (still) requires rethinking generalization, Communications of the ACM, Vol. 64, No. 3, 2021. [7] G. B. Moody, and R. G. Mark, A database to support development and evaluation of intelligent intensive care monitoring, Computers in Cardiology, 1996. [8] G. Huang, Z. Liu, L. V. D. Maaten, and K. Q. Weinberger, Densely connected convolutional networks, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. [9] H. I. Fawaz, G. Forestier, J. Weber, L. Idoumgharand, and P. A. Muller, Deep learning for time series classification: a review, Data Mining and Knowledge Discovery, Vol 33, 2019. [10] H. K. Choi, Stock price correlation coefficient prediction with arima-lstm hybrid model, arXiv preprint arXiv:1808.01560, 2018. [11] H. Li, Multivariate time series clustering based on common principal component analysis, Neurocomputing, Vol 349, 2019. [12] J. Y. Syu, 美股大跌歷史回顧：20世紀百年來美國股市的7次重大股災, Web Page, https://rich01.com/historical-stock-crush-20-century/?fbclid=IwAR0cQ4E YAj02o1Rc0Q3d5_5kMOFiUm3bSOphtCqPmvJ4W6RLbhePBtLKKFY#%E7%BE%8E%E8%82%A1%E8%82%A1%E7%81%BD6%EF%BC%9A2007%E5%B9%B4%E6%AC%A1%E8%B2%B8%E5%8D%B1%E6%A9%9F, 2022 [13] K. He, X. Zhang, S. Ren, and J. Sun, Deep residual learning for image recognition, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. [14] K. Senneset and M. Gultvedt, Something old, something new: a hybrid approach with arima and lstm to increase portfolio stability, Master Thesis, NHH Norwegian School of Economics, 2020. [15] S. Du, T. Li, Y. Yang, and S. J. Horng, Multivariate time series forecasting via attention-based encoder–decoder framework, Neurocomputing, Vol. 388, 2020. [16] S. Han and S. S. Woo, Learning sparse latent graph representations for anomaly detection in multivariate time series, Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022. [17] S. Raschka, Understanding and Coding Self-Attention, Multi-Head Attention, Cross-Attention, and Causal-Attention in LLMs, Web Page, https://magazine. sebastianraschka.com/p/understanding-and-coding-selfattention, 2024. [18] S. Tuli, G. Casale and N. R. Jennings, Tranad: deep transformer networks for anomaly detection in multivariate time series data, Proceedings of the VLDB Endowment, Vol 15, No. 6, 2022. [19] S. Y. Shih, F. K. Sun, and H. Y. Lee, Temporal pattern attention for multivariate time series forecasting, Machine Learning, Vol 108, 2019. [20] T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie, and M. Li, Bag of tricks for image classification with convolutional neural networks, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 描述 碩士
國立政治大學
資訊科學系
110753201資料來源 http://thesis.lib.nccu.edu.tw/record/#G0110753201 資料類型 thesis dc.contributor.advisor 沈錳坤 zh_TW dc.contributor.advisor Shan, Man-Kwan en_US dc.contributor.author (Authors) 曹昱維 zh_TW dc.contributor.author (Authors) Tsao, Yu-Wei en_US dc.creator (作者) 曹昱維 zh_TW dc.creator (作者) Tsao, Yu-Wei en_US dc.date (日期) 2024 en_US dc.date.accessioned 1-Mar-2024 13:42:40 (UTC+8) - dc.date.available 1-Mar-2024 13:42:40 (UTC+8) - dc.date.issued (上傳時間) 1-Mar-2024 13:42:40 (UTC+8) - dc.identifier (Other Identifiers) G0110753201 en_US dc.identifier.uri (URI) https://nccur.lib.nccu.edu.tw/handle/140.119/150173 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 資訊科學系 zh_TW dc.description (描述) 110753201 zh_TW dc.description.abstract (摘要) 多變量時間序列的研究包括預測、分類、分群和異常偵測等。其中，時間序列異常偵測是近年熱門的研究主題。其目標在早期發現時間序列中的異常現象。針對多變量時間序列的相關性異常偵測，很少現有研究。本研究的主要目的是偵測多變量時間序列資料的相關性異常。 本研究採取非監督式學習，根據正常多變量序列所學習的相關性預測模型，來偵測相關性異常。本研究提出 CARG模型 (Clustering-Attention-Residual-GRU Model)來預測多變量時間序列之間的相關性。CARG 模型不僅能夠捕捉相關性本身的時間性結構，同時還透過學習多變量之間的相依性資訊，從而提高預測效果。CARG 模型結合了時間序列分群、注意力機制和時間序列分析模型等技術，具有能夠有效捕捉多變量之間相依性結構的優勢。當 CARG 模型的預測值與實際值存在顯著差異時，可能發生相關性異常事件。 在我們的實驗中，我們將 CARG 模型應用於金融市場資料上，CARG 模型展現出相當不錯的表現。此外，實驗結果也表明，在 CARG 模型中所設計的各結構對多變量時間序列相關性預測任務均有所助益。並且透過與Baseline模型的比較，CARG 模型確實加強預測的效能。 zh_TW dc.description.abstract (摘要) The research domain of multivariate time series consists of forecasting, classification, clustering, and anomaly detection, among others. Anomaly detection in time series has become a popular research topic in recent years. Its goal is to discover anomalies in time series data at an early stage. Little research has been paid on the anomaly detection of correlations in multivariate time series. The primary objective of this thesis is to detect correlation anomalies in multivariate time series data. This research adopts unsupervised learning approach to detect correlation anomalies based on a correlation prediction model learned from normal multivariate series. We propose the CARG model (Clustering-Attention-Residual-GRU model) to predict the correlations among multivariate time series. The CARG model not only captures the temporal structure of the correlations but also improves prediction performance by learning the dependency information among the multivariates. Combining techniques of time series clustering, attention mechanisms, and time series analysis models, the CARG model is advantageous in effectively capturing the dependency structure among multivariates. To detect correlation anomalies, a significant discrepancy between the predicted values by the CARG model and the actual values may indicate a correlation anomaly event. In our experiments, we applied the CARG model to financial market data and it showed quite impressive performance. Additionally, the experimental results also demonstrate that the various structures designed in the CARG model contribute positively to the task of predicting correlations in multivariate time series. By comparison with baseline models, the CARG model indeed enhances the effectiveness of prediction. en_US dc.description.tableofcontents 致謝 i 摘要 ii Abstract iii 目錄 v 表目錄 vi 圖目錄 vii 1. 緒論 1 1.1 研究背景 1 1.2 研究動機 2 1.3 研究目的 2 2. 相關研究 4 2.1 時間序列的異常偵測 4 2.2 時間序列相關性預測 6 3. 研究方法 8 3.1 模型流程概述 8 3.2 Correlation Transformation 8 3.3 Clustering 11 3.4 Prediction Model 15 3.6 Anomaly Estimation 25 4. 實驗設計與結果 27 4.1 實驗目的 27 4.2 資料集 27 4.3 評估指標 28 4.4 實驗設計與結果 29 5. 結論 44 參考文獻 46 zh_TW dc.format.extent 2066670 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0110753201 en_US dc.subject (關鍵詞) 多變輛時間序列 zh_TW dc.subject (關鍵詞) 相關性預測 zh_TW dc.subject (關鍵詞) 相關性異常偵測 zh_TW dc.subject (關鍵詞) Multivariate Time Series en_US dc.subject (關鍵詞) Correlation Prediction en_US dc.subject (關鍵詞) Correlation Anomaly Detection en_US dc.title (題名) 時間序列資料的相關性異常偵測 zh_TW dc.title (題名) Anomaly Detection of Correlation Change over Multivariate Time Series en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) [1] A. Blázquez-García, A. Conde, U. Mori, and J. A. Lozano, A review on outlier/anomaly detection in time series data, ACM Computing Surveys, Vol. 54, No. 3, 2021. [2] A. Deng, and B. Hooi, Graph neural network-based anomaly detection in multivariate time series, Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35, No. 5, 2021. [3] A. Hanni, Correlation-based anomaly detection in time series, Master Thesis, Department of Informatics, Universiy of Bern, 2020. [4] A. L. Goldberger, et al., Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals, Circulation, Vol. 101, No. 23, 2000. [5] A. Vaswani, et al., Attention is all you need, Advances in Neural Information Processing Systems (NIPS), Vol. 30, 2017. [6] C. Zhang, S. Bengio, M. Hardt, B. Recht, and O. Vinyals, Understanding deep learning (still) requires rethinking generalization, Communications of the ACM, Vol. 64, No. 3, 2021. [7] G. B. Moody, and R. G. Mark, A database to support development and evaluation of intelligent intensive care monitoring, Computers in Cardiology, 1996. [8] G. Huang, Z. Liu, L. V. D. Maaten, and K. Q. Weinberger, Densely connected convolutional networks, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. [9] H. I. Fawaz, G. Forestier, J. Weber, L. Idoumgharand, and P. A. Muller, Deep learning for time series classification: a review, Data Mining and Knowledge Discovery, Vol 33, 2019. [10] H. K. Choi, Stock price correlation coefficient prediction with arima-lstm hybrid model, arXiv preprint arXiv:1808.01560, 2018. [11] H. Li, Multivariate time series clustering based on common principal component analysis, Neurocomputing, Vol 349, 2019. [12] J. Y. Syu, 美股大跌歷史回顧：20世紀百年來美國股市的7次重大股災, Web Page, https://rich01.com/historical-stock-crush-20-century/?fbclid=IwAR0cQ4E YAj02o1Rc0Q3d5_5kMOFiUm3bSOphtCqPmvJ4W6RLbhePBtLKKFY#%E7%BE%8E%E8%82%A1%E8%82%A1%E7%81%BD6%EF%BC%9A2007%E5%B9%B4%E6%AC%A1%E8%B2%B8%E5%8D%B1%E6%A9%9F, 2022 [13] K. He, X. Zhang, S. Ren, and J. Sun, Deep residual learning for image recognition, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. [14] K. Senneset and M. Gultvedt, Something old, something new: a hybrid approach with arima and lstm to increase portfolio stability, Master Thesis, NHH Norwegian School of Economics, 2020. [15] S. Du, T. Li, Y. Yang, and S. J. Horng, Multivariate time series forecasting via attention-based encoder–decoder framework, Neurocomputing, Vol. 388, 2020. [16] S. Han and S. S. Woo, Learning sparse latent graph representations for anomaly detection in multivariate time series, Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022. [17] S. Raschka, Understanding and Coding Self-Attention, Multi-Head Attention, Cross-Attention, and Causal-Attention in LLMs, Web Page, https://magazine. sebastianraschka.com/p/understanding-and-coding-selfattention, 2024. [18] S. Tuli, G. Casale and N. R. Jennings, Tranad: deep transformer networks for anomaly detection in multivariate time series data, Proceedings of the VLDB Endowment, Vol 15, No. 6, 2022. [19] S. Y. Shih, F. K. Sun, and H. Y. Lee, Temporal pattern attention for multivariate time series forecasting, Machine Learning, Vol 108, 2019. [20] T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie, and M. Li, Bag of tricks for image classification with convolutional neural networks, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019. zh_TW
