Time Series Anomaly Detection and Prediction Model Integrating Multimodal Data

Time series anomaly detection in multimodal data is crucial for data analysis and mining. However, existing multimodal detection methods mainly focus on single-mode data and fail to effectively explore the distribution characteristics of multimodal data in multimodal space, the adaptive fusion of multimodal features, and the extraction of spatiotemporal dependencies between multimodal features. To address the above issues, this project aims to study a multi-modal feature adaptive fusion method based on one-dimensional convolutional networks, in order to solve the problem of adaptive fusion between multi-source temporal and multi-modal features. For the fused multimodal features, a spatiotemporal attention model can be established that includes both temporal and spatial attention dimensions, and its spatiotemporal correlation can be extracted to obtain spatiotemporal attention vectors, thereby obtaining prediction results. Through testing on three public datasets, the results show that the time series anomaly detection method based on multimodal feature fusion has significant advantages.