<p>To enhance the reliability of cable force optimization in large-span cable-stayed bridges, this study presents a force optimization model that considers reliability indicators specific to these types of bridges. A structural surrogate model was established by employing a Radial Basis Function Neural Network (RBFNN) to accurately capture the mapping relationship between random variables and the structural response. Enhancements were introduced to address the limitations of the standard Seagull Optimization Algorithm (SOA) through refracted backpropagation learning and nonlinear convergence strategies. A combined force optimization method was devised by integrating the RBFNN and the improved SOA. An empirical analysis was performed on a large-span cable-stayed bridge to validate the feasibility of the proposed approach. The results demonstrated the RBFNN’s ability to effectively capture the nonlinear mapping between structural random variables and dynamic responses. The enhanced seagull algorithm exhibited substantial performance improvements compared to the original algorithm, providing better solutions for force optimization considering reliability indicators. Following optimization, although the overall trend of tension distribution remained similar to the original distribution, adjustments were made to specific tension points to varying degrees. Notably, the deflection of the main beam in the middle span was significantly improved, with a maximum reduction of approximately 36.21%. Furthermore, there was a slight improvement in the reliability indicators for tension, with a maximum increase of approximately 9%.</p>