一种面向时间序列分类的门控多尺度时空图卷积模型OA

Time series classification is a significant research direction in the fields of data mining and machine learning,with extensive applications across various domains.However,as the scale and complexity of time series data have increased substantially,efficiently capturing temporal dependencies and feature interactions within the data has become a core challenge in time series classification tasks.In recent years,an increasing number of researchers have attempted to apply graph neural networks(GNNs)to time series data that does not inherently exhibit explicit graph structures.Nevertheless,limitations remain in the construction of graph structures and the representation of the sequential nature of time series.To address these issues,a gated multi-scale spatio-temporal graph convolutional network(GMS-STGCN)is proposed for time series classification tasks.GMS-STGCN integrates multi-scale features of time series with deep learning models and graph neural networks,leveraging GNNs to extract correlations and differences among features across different scales and time slots.Additionally,a novel multi-scale feature fusion mechanism and a progressive gated propagation mechanism are designed to further extract temporal features,culminating in the integration of an attention mechanism for comprehensive feature extraction.Experimental results demonstrate that GMS-STGCN achieves superior performance on multiple public time series classification datasets,validating its effectiveness in feature interaction modeling,multi-scale feature capturing,and dynamic feature dependency modeling.