Hybrid Physics–AI Earthquake Forecasting Using a Dynamic K–R Nonlinear Instability Framework

This work presents a hybrid nonlinear framework for earthquake instability modeling and forecasting based on a dynamic excitation–regulation (K–R) formulation. The proposed approach integrates tectonic stress evolution, energy redistribution, clustering behavior, and system memory with data-driven predictive modeling to represent earthquake occurrence as a state-dependent nonlinear process. The framework models seismic systems as evolving instability fields in which excitation reflects stress and energy accumulation, while regulation represents stabilization through redistribution and temporal memory effects. Machine learning operates within this physically informed state space to estimate earthquake likelihood and detect transitions toward rupture conditions. Experimental evaluation using seismic catalog–derived datasets demonstrates strong discrimination capability and robustness across multiple scenarios, including lead-time forecasting, major earthquake detection, temporal generalization, and noise sensitivity analysis. The hybrid KR-based system achieves high predictive performance while maintaining physical interpretability relative to purely statistical or data-driven models. This preprint represents a research manuscript prepared for submission to a peer-reviewed journal. The posted version is a non–peer-reviewed draft intended for academic dissemination and feedback.