基于神经网络和演化算法的土石坝位移反演分析

构造了基于综合应用人工神经网络和演化算法的位移反演分析方法。该法使用具有较强非线性映射能力的神经网络模型代替有限元计算，提高了计算效率。采用演化算法和Vogl快速算法，同时优化神经网络的结构和权值，增加其适应性并加快训练速度；使用多种群演化等策略，改善演化算法的全局收敛性和收敛速度。以三峡茅坪溪防护土石坝的变形反演分析为例，研究了神经网络演化代数以及训练样本数量对神经网络模拟能力的影响，证明了所建立的反演分析方法的有效性。

Displacement back analysis of embankment dam based on neural network and evolutionary algorithm

A new approach of displacement back analysis is proposed by combining the neural network and evolutionary algorithm. The neural network with optimal architecture trained by evolutionary algorithm and Vogl algorithm is used to substitute the time-consuming finite element analysis. The convergence of search is improved and speeded up by evolution strategies such as multi-population. The proposed approach is verified by applying it to the displacement back analysis of Maopingxi embankment dam in Three Gorges Project: and the influence of generation number and sample size on the simulation ability of neural network is studied.