进化算法结合神经网络反演刚性路面模量

Module backcalculating of rigid pavement with artificial neural networks and genetic algorithm

为解决传统路面结构参数反演分析方法易陷于局部最优解的问题,根据弹性地基上的小挠度薄板理论,建立了刚性路面有限元计算模型,模型中假定接缝只传递剪力,用压缩柔度矩阵方法考虑接缝的约束条件。用有限元模型构造训练集对BP神经网络进行训练,依靠BP网络的记忆和联想功能建立结构层模量与路面弯沉盆之间的映射关系,利用进化算法(GA)在搜索空间内找出使弯沉盆误差最小的一组模量组合。实例计算结果表明进化算法结合神经网络方法反演刚性路面模量的最大误差只有2.8％。

On the ground of small displacement slab above elastic foundation, a finite element model was firstly developed for calculating displacement of rigid pavements with boundary effect of joints taken into account. The BP neural networks was then trained with the training congregate built from rigid pavement model to establish the relationship of module and surface deflection basin. A method was developed for pavement module backcalculating with genetic algorithm simulating the evolution of nature. The best group of module was searched at random within entire searching space to satisfy the minimum error request of deflection basin. The GA-ANN method is an efficiency approach for module back calculation. It can be seen from the numerical results that the maximal error of backcalculating module is 2.8% only.