Parameter calibration of finite element model of reinforced concrete arch bridge based on ISOA-RBF neural network

To achieve efficient and accurate finite element (FE) model updating for reinforced concrete (RC) arch bridges, this study proposes a novel methodology that integrates an Improved Seagull Optimization Algorithm (ISOA) with a Radial Basis Function (RBF) neural network. First, the base Seagull Optimization Algorithm (SOA) is enhanced through Logistics chaotic mapping and an adaptive attack strategy, forming the ISOA to improve search efficiency and convergence. Second, leveraging the superior nonlinear fitting capability of the RBF neural network, a surrogate model is constructed to map the relationship between key structural parameters and the corresponding static/dynamic responses of the bridge, thereby replacing the computationally expensive full FE model during iterative optimization. Finally, a multi-objective optimization model is formulated by combining static deflections (from arch T-beams) and modal frequencies, enabling a holistic FE model update based on in-situ load test data. The proposed ISOA-RBF framework is applied to a long-span RC arch bridge case study. Results demonstrate that: (1) the ISOA effectively solves the inverse optimization problem in FE model updating; (2) the RBF surrogate model achieves high approximation accuracy, with an average relative error of only 0.62% across 50 validation samples; (3) after updating, discrepancies between computed and measured T-beam deflections are significantly reduced, showing average relative errors below 10% under two static load cases; (4) errors in the first three modal frequencies are less than 5%, with good orthogonality maintained among modes and no modal mixing observed, confirming consistency in dynamic characteristics before and after model updating.