DSPR: Dual-Stream Physics-Residual Networks for Trustworthy Industrial Time Series Forecasting

In three linesDSPR (Dual-Stream Physics-Residual Networks) proposes a forecasting framework that decouples stable temporal patterns from regime-dependent residual dynamics in industrial time series. Using an Adaptive Window module and Physics-Guided Dynamic Graph, it achieves 99% Mean Conservation Accuracy and 97.2% Total Variation Ratio across four industrial benchmarks.

Summary generated by Claude — human-verified