基于RBF神经网络的软基沉降预测研究

Research of settlement prediction of soft foundation based on RBF neural networks

将神经网络理论引入软基沉降预测领域.借助自控领域信号处理的思想,应用改进后的径向基函数神经网络的映射模式进行软基沉降的短期预测;软基沉降的长期预测实质上为基于神经网络的多维欧氏空间的曲面拟合问题,将地基压缩层从上到下分成若干段,每段的土性指标按段内各层土在段中的长度取加权平均作为系统的输入,将某个沉降模型的沉降曲线参数作为系统的输出,可以预测后期沉降曲线走势.实践表明,建立的基于RBF神经网络的软基沉降短期预测和长期预测模型是可行的,只要有足够多的训练样本,长期预测可以达到比较精确的预测效果.表5,参9.

The neural networks theory is applied to predicting soft foundation setdement field. Shortdated prediction uses improved mapping mode of Radial Basis Function in virtue of signal processing at autocontrol scopes. Long - term prediction of soft foundation settlement is virtually a problem of Euclid＇ s multidimensional Euclid space surface fitting, the input data is weighted mean of soll mechanics parameter, and the output is settlement parameter of certain model, so we can predict the trend of settlement curve. Testified by practice, the shortdated prediction and long - term prediction of soft foundation settlement based on RBF（ Radial Basis Fanction）neural networks is feasible. Having enough sample, the long- term prediction is able to receive accurate result. 5tabs. ,9refs.