Multidimensional time series clustering is an important problem in time series data analysis. This paper provides a new research idea for the behavioral analysis of financial markets, using the intrinsic correlation existing between transactions in the same segment of the financial market to cluster and analyze multidimensional time-series data, so as to obtain different types of market characteristics. In this paper, we propose a multidimensional time series clustering model based on graph attention autoencoder (GATE) and mask self-organizing map (Mask-SOM), based on which we realize multi-step prediction of financial derivatives prices and intelligent trading system construction. To obtain and fully utilize the correlation features between multidimensional financial time series data containing high noise for clustering analysis, constant curvature Riemannian manifolds are introduced in the graph attention autoencoder, and the multidimensional financial time series features captured by the encoder are embedded into the manifold. Following that, the multidimensional financial time series clustering analysis is implemented using Mask-SOM analysis manifold encoding. Finally, the feasibility and effectiveness of the model are verified using real financial datasets.
翻译:在时间序列数据分析中,多层面时间序列群集是一个重要问题。本文件为金融市场行为分析提供了一个新的研究理念,利用金融市场同一部分交易之间的内在关联,对多层面时间序列数据进行分组和分析,以获得不同类型的市场特征。在本文中,我们提出了一个基于图形注意自动编码器(GATE)和掩码自我组织地图(Mask-SOM)的多层面时间序列群集模型,我们在此基础上对金融衍生物价格和智能贸易系统建设进行多步预测。为了获得并充分利用包含高噪音的多维金融时间序列数据之间的关联性,在图形注意中引入了恒定的里曼式数字元件,而编码器所捕捉的多层面财务时间序列特征嵌入了元件中。随后,利用Mask-SOM分析的多重编码进行了多层面财务时间群集系列分析。最后,利用真实的财务数据集验证了模型的可行性和有效性。