Autocorrelation-based Signal Reconstruction for Multi-source time series Anomaly Detection

In the quest to enhance IoT device performance and expand cross-domain system functionalities, pinpointing anomalies in multi-source time series data becomes crucial. Traditional deep learning approaches often overlook the complex patterns in data. Our innovative method integrates time series decomposition into anomaly detection, thus revealing intrinsic periodicities vital for advanced engine and powertrain modeling. By employing a joint loss function, our approach not only improves anomaly detection accuracy and stability but also aligns with essential aspects of interconnected and automated vehicle control. Our method, proven superior through extensive comparison across datasets and algorithms, significantly advances engine and powertrain supervision, management, diagnostics, and modeling, showcasing a notable contribution to these fields.