基于神经网络和遗传算法的桥梁参数优化方法与分析

在桥梁施工监控中,实测标高与理论预测标高存在差异的主要原因之一是理论计算参数存在偏差,因此对计算参数进行有效修正成为现代施工控制的一个关键问题。分析和确定了影响标高预测误差的主要参数,利用有限元分析建立了主要参数和标高之间的BP网络模型,该模型在一定参数取值范围内可以取代有限元模型预测标高,从而大大减少计算量。在建立的BP模型基础上,通过浮点编码的遗传算法对目标函数进行优化得到一组最优计算参数。对实际桥梁的计算分析表明,本文确定的分析参数物理意义明确,基于本文方法的修正能有效地提高标高预测精度,对类似桥梁的计算分析具有指导意义。

Optimization of Bridges'Parameters Based on BP Neural Network and Genetic Algorithm

During the construction control of bridges,one of the main causes about the differences between the measured elevation and the predicted elevation is that the parameters in the computation of the predicted elevation are not sufficiently identical with the practical ones.Therefore the effective correction of the computational parameters becomes a key issue in modern construction control.In this paper,the key parameters affecting the elevation differences were analyzed and determined first.A BP neural network model based on the FEM analysis was set up relating the elevation to the key parameters,which may substitute the FEM to predict the elevation when the parameters take value in given ranges.This can reduce the computational complexity significantly.Based on the BP model,a floating-point encoding genetic algorithm was used to optimize objective function for the optimized key parameters.An example from a practical engineering problem showed that the determined key parameters have clear physics meanings.Their corrections based on the suggested methods in the paper can enhance the prediction accuracy of the elevations effectively and the optimized methods recommended in the paper could be good reference for the analysis of similar bridges.