边坡岩体力学参数反分析遗传—神经网络算法

A Genetic-artificial Neural Network Algorithm for Mechanical Parameters of Rock Slope

基于均匀设计、有限元法、人工神经网络和遗传算法建立了新的边坡岩体力学参数反分析方法。按照均匀设计要求,确定数值模拟方案;用有限元程序计算出相应的神经网络训练样本,建立边坡变形的神经网络预测模型,再利用遗传算法进行反演分析,其中反演过程适应度的计算则采用已训练好的神经网络预测来替代有限元数值仿真,这样大大缩短了计算时间。通过算例分析,反演结果比较理想,表明该反分析方法是可行性和精确的。

A new back analysis method for mechanical parameters of slope rocks is developed based on uniform testing design, finite element method, artificial neural network and genetic algorithm. According to uniform testing design, the value levels of the mechanical parameters are chosen, and simulation schemes are arranged; the related analytical samples for neural network are given by FEM calculations. Thus, a BP neural network which is used to forecast displacement of the slope＇ s character points is erected and trained. The physical and mechanical parameters can be analyzed backwards by genetic algorithm. In this algorithm the trained BP neural network is used to calculating the fitness value instead of the FEM method and the calculation time is much reduced. Through examples analysis, the error between the back analysis results and the theoretical ones is much less and meets the requirement of precision, which indicates that this back analysis method is feasible and accurate.