Zar-Manah HELIOS 2.0: Open-Source Artificial Intelligence (AI) Models for Real-Time Solar Storm Prediction and Early Warning Across Solar Cycles 23–25
Year: 2026
Type: preprint
Abstract: HELIOS 2.0 is an open-source suite of Artificial Intelligence (AI) models engineered to predict dangerous solar storms, coronal mass ejection (CME) shock arrival times, and geomagnetic blackouts from live satellite telemetry in real time. Operating as a synchronized 4-gear cascaded AI pipeline executing in <25 ms on consumer hardware (Apple Silicon M4 GPU), HELIOS 2.0 eliminates the multi-hour supercomputing bottlenecks of legacy numerical MHD simulations (WSA-ENLIL) and eliminates 92.1% of operational false alarms: Gear 1 (Solar Coronal Chirality AI - HELIOS-VISION-APEX): Deep convolutional neural network (MobileNetV3) analyzing SOHO/LASCO C2 coronagraph imagery and SDO/HMI magnetograms (Schrijver R-value) to predict magnetic cloud southward polarity (Bz ≤ -10 nT) 24–72 hours prior to arrival. Gear 2 (Storm Severity AI Veto - HELIOS-STORM-APEX): AI severity classifier predicting ring current collapse (Dst ≤ -110 nT) with 87.0% out-of-distribution accuracy and 0.8717 ROC-AUC across Solar Cycle 25, defusing false alarms caused by non-geoeffective fast CMEs. Gear 3 (3D Interplanetary Kinematics AI - MoKTransitNN & HTWS): Dynamic Mixture-of-Kinematics neural network with closed-form Weight-Space Surgery (HTWS) and aerodynamic drag enforcement (transit ≥ 16.0h), achieving a median arrival error of ±3.42 hours across historical events and beating NASA WSA-ENLIL. Gear 4 (L1 In-Situ Attention AI - HeliosBiLSTM): 24-hour multivariate temporal attention network monitoring real-time DSCOVR/ACE solar wind plasma at L1 for nowcasting (+1h, +2h, +3h, +6h) with mathematically guaranteed conformal confidence bands. This package contains the formal 3-page scientific preprint (PDF), the interactive browser-based mission cockpit (Zar-Manah_HELIOS.html) featuring zero-disk volatile RAM streaming, trained PyTorch and LightGBM models, and evaluation benchmarks.
Author: Zar Manah
Language: English
Subfield: Astronomy and Astrophysics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225576262