基于GA-ANN算法的层状路基土参数预测模型

A Parameter Prediction Model for Layered Foundation Soil Based on GA- ANN

将BP神经网络和遗传算法两种智能方法结合起来,建立起自适应遗传算法-BP神经网络系统。利用土工试验得到的不同土层物理力学参数汇总整理形成的试验数据作为样本值,对路基土层物理力学参数进行了预测,将预测结果和单独使用BP神经网络时的预测结果进行了对比分析。结果表明:当样本数据离散性小时,这两种预测方法均能取得理想的预测效果,自适应遗传算法-BP神经网络系统还具有有效防止"过训练"和提高网络自身的泛化能力;当样本规模大,且样本数据具有一定的离散性时,网络系统的预测优势能更好地体现出来。

In this paper ,the two intellectual technologies of BP neural networks and genetic algorithm are considered and banded together to establish the self-adaptive genetic algorithm and BP neural network system used to predict the parame-ters for layered soil .Lots of physical and mechanical parameters of different layered soils obtained from experiments in soil mechanics laboratory are sorted out and used for the sample of the system ,then ,the target parameters of layered soil are predicted by the system .A comparison analysis is conducted between the two kinds of prediction results with the sys-tem mentioned above and BP neural networks respectively .It shows that the ideal prediction results can be obtained si-multaneously by the two methods while the variance of the sample data is small .The self-adaptive genetic algorithm and BP neural network system can also provide the generalization function to prevent the “overfull training”;When the sample scale and variance of sample data are both big enough ,the superiority of the network system can be better expressed .