Traditional bridge design, centralizing seismic safety while neglecting post-earthquake restoration efficiency, was struggled to meet seismic resilience needing. This study proposed a multi-objective optimization method for enhancing the seismic resilience of High-Speed Railway Track Bridge Systems (HSRTBS), which integrated both system fragility and restoration efficiency by machine learning techniques. A structural surrogate model was developed using finite element simulations and machine learning algorithms to predict earthquake responses of railway bridge. A multi-objective optimization model, selected equivalent downtime of components and the resilience index as objective functions, was formulated and solved using NSGA-II, NSGA-III, and MOEA\D. In this paper, multi-objective optimization design of seismic resilience of the 8-span simply supported girder bridge with a length of 32 m was investigated. The results indicated that Random Forest, AdaBoost, Extra Trees, and LightGBM were satisfactory for structural response prediction of HSRTBS, with ASI, SMA, and PP parameters shown high contribution levels in predictions. The fragility analysis results revealed that the damage sequence of components from the sliding layer to the fixed bearing, shear slots, and finally the piers, while the sliding bearing, CA mortar layer, fasteners, and shear rebar remained undamaged or slight damage. The key components that control the restoration performance of the system were the fixed bearing and shear slots. By appropriately increasing the shear strength of shear slots and shear rebar, it can still achieve the seismic resilience of the nonoptimized pier with the 15.61 % reduction in the comprehensive reinforcement quantity of bridge piers. This study explores a new way to enhance seismic resilience by optimizing the reinforcement ratio of railway bridge piers.

Multi-objective optimization;Machine learning;High-speed railway track bridge system;Seismic resilience;Restoration efficiency