一种基于GA-ANN算法的层状土参数预测模型

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

结合BP神经网络和遗传算法,建立起自适应遗传算法-BP神经网络系统.以不同土层大量物理力学参数汇总整理形成的试验数据作为样本值,应用该系统对地基土层物理力学参数进行了预测,并将预测结果和BP神经网络的预测结果进行对比分析.结果表明:当样本数据离散性小时,两种方法均能取得理想的预测效果,而且所建立的系统还能有效防止"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,该网络系统的预测优势能更好地体现出来.

BP neural networks and genetic algorithm were combined together to estamlsn me sen- adaptive genetic algorithm and BP neural network system, and were used to predict the parameters of layered soil. Lots of physical and mechanical parameters of different layered soils are sorted out and used as the sample, then, the target parameters of layered soil were predicted. The results predicted with two kinds of intellectual technologies were compared with those predicted with BP neural networks. It shows that the ideal prediction results can be obtained simultaneously by the two technologies while variance of the sample data is small. The established system can also provide itself the generalization function to prevent the case ＂overfull training＂. When sample scale and variance of sample data are both big enough, the superiority of the network system can be better expressed.