PISCES: Physics-Informed Solar-wind Convolutional AutoEncoder for Space-weather Anomaly Detection and Early Warning

PISCES is a convolutional autoencoder architecture designed for space weather anomaly detection and early warning. The model is trained on OMNI solar wind measurements without relying on catalog labels, incorporating physics constraints to improve detection accuracy. It targets solar wind transients at the Sun-Earth first Lagrange point (L1), before these events reach Earth, addressing the limitations of fixed-threshold methods that may miss combined magnetic and plasma structures. The approach provides a novel solution for detecting complex space weather phenomena.