Fall accidents among older adults represent a major public health challenge worldwide. Energy-absorbing flooring has gained increasing attention due to its high usability and robustness against fall-related hip impacts. This study proposes a novel method that integrates finite element (FE) analysis, deep learning(DL) models, and multi-objective optimization (MOO) algorithms to enhance the biomechanical protective performance of a bio-inspired energy-absorbing structure. To achieve this, 100 structural configurations were generated based on a design of experiments (DOE) framework, automatically modeled in Hypermesh, and integrated into a hip regional model in LS-DYNA. The deep neural network (DNN) models were developed to predict femoral neck force (𝐹𝑛𝑒𝑐𝑘) and energy absorption efficiency (𝑆𝐸𝐴𝑣), and were subsequently utilized in the MOO framework to construct the Pareto front, optimizing the dual objectives of minimizing 𝐹𝑛𝑒𝑐𝑘 and maximizing 𝑆𝐸𝐴𝑣 using 50,000 optimization samples. Five optimal solutions on the Pareto front were validated and demonstrated substantial performance improvements. Compared to the baseline structure, the optimized designs demonstrated up to a 23% reduction in peak femoral neck force and a 65% increase in energy absorption efficiency. This study presents a framework that ensures high accuracy, robustness, and continuity in representing biomechanical responses towards improved hip protection. The findings have practical implications for enhancing safety in high fall-risk environments and provide valuable guidance for manufacturers in designing protective devices with enhanced performance and clinical effectiveness.