Integrating ML-based reverse engineering of CGE models with optimization for enhancing resilience

Abstract

Natural hazards pose significant risks to economic stability and community well-being, particularly in regions vulnerable to seismic activity. This research develops a novel framework integrating machine learning (ML) and optimization techniques to enhance community resilience against natural hazards, with a specific focus on earthquake impacts. A Computable General Equilibrium (CGE) model is employed to simulate economic outcomes, such as domestic supply, income, employment, and migration, under various earthquake scenarios. To overcome the computational challenges of CGE models, an Elastic Net regression-based ML model is utilized as a surrogate, achieving a predictive accuracy of over 98% (R2). These predictive insights are incorporated into an integer linear programming (ILP) optimization model to identify optimal retrofit strategies for minimizing economic loss while addressing resource constraints.The framework is applied to Salt Lake City (SLC), Utah, a region highly vulnerable to seismic activity. Using 5,000 earthquake simulations, the approach evaluates economic impacts across 11 sectors and seven geographic regions. The optimization model ensures equitable resource allocation and effectively prioritizes infrastructure retrofits based on sector-specific vulnerabilities and budget limitations. This work makes significant contributions by: (i) introducing a scalable framework that integratesCGE models, ML, and optimization for disaster resilience, (ii) demonstrating the effectiveness of Elastic Net regression in balancing model accuracy and interpretability, and (iii) providing actionable insights for policymakers to enhance disaster preparedness and recovery strategies. The findings emphasize the value of advanced computational tools in informing data-driven decision-making for community resilience. Future research will focus on extending the methodology to multi-hazard scenarios and dynamic resilience metrics to broaden its applicability and impact.