Mathematics, Год журнала: 2024, Номер 12(21), С. 3320 - 3320

Опубликована: Окт. 23, 2024

This paper introduces Tensor Visibility Graph-enhanced Attention Networks (TVGeAN), a novel graph autoencoder model specifically designed for MTS learning tasks. The underlying approach of TVGeAN is to combine the power complex networks in representing time series as graphs with strengths Graph Neural (GNNs) from data. consists two new main components: TVG which extend capabilities visibility algorithms MTSs by converting them into weighted temporal where both nodes and edges are tensors. Each node represents observations at particular time, while weights defined based on angle algorithm. second component proposed GeAN, attention mechanism developed seamlessly integrate interactions represented core process. GeAN achieves this using outer product quantify pairwise fine-grained level bilinear effectively distil knowledge interwoven these representations. From an architectural point view, builds complemented sparse variational units. unit used promote inductive TVGeAN, endow generative capabilities. performance extensively evaluated against four widely cited benchmarks supervised unsupervised results evaluations show high various In particular, can achieve average root mean square error 6.8 C-MPASS dataset (i.e., regression tasks) precision close one SMD, MSL, SMAP datasets anomaly detection tasks), better than most published works.

Язык: Английский