Improved-GEP-based models for predicting distortion control indices of curved composite box girders

Abstract

Curved steel-concrete composite box girders are subjected to complex torsional actions under vertical loads, which may generate excessive distortional warping stresses. However, the complexity of the distortional warping stresses makes it challenging to obtain a quantitative prediction method, which may result in the adoption of potentially unsafe or uneconomical diaphragm designs. In this paper, improved-GEP (gene expression programming)-based models for predicting two distortion control indices of curved composite girder bridges are proposed. The key parameters in distortion equation for the curved composite box girder bridges are derived first, and then a comprehensive parametric analysis is conducted. Subsequently, an improved GEP algorithm is proposed to establish the formulas for calculating two distortion control indices. Compared to the traditional GEP, the proposed algorithm can markedly improve the mining ability of the constant terms and coefficients and enhance the global search ability by introducing a least squares algorithm and a new-generation and quasi-offspring mixed competitive strategy. In addition, with the operative and terminal symbols determined based on distortion theory, the efficiency and physical significance of the GEP models are effectively enhanced. The statistical indices and sensitivity analysis demonstrate that the proposed models for predicting the two distortion control indices agree with the finite element analysis results and distortion theory.