滑带黄土振陷预测中的BP神经网络方法

The method of BP neural network in slip zone loess seismic subsidence forecast

目的针对陇海铁路沿线密集的黄土滑坡灾害,对其振陷系数提出一种新的预测方法。方法以滑带黄土动三轴试验资料为基础,运用MATLAB建立滑带黄土振陷的BP(error back propagationneural network)神经网络预测模型,并与多元线性回归方法建立的模型进行误差对比分析。结果BP神经网络模型预测的结果要比多元线性回归模型预测的更准确。结论滑带黄土振陷预测的BP神经网络模型是一种比较理想的预测方法,对黄土地区的滑坡稳定性评价和铁路地基沉降的分析具有重要的价值。

Aim To present a new method for predicting seismic subsidence coefficient of dense landslides along Longhai railway.Methods To establish a BP neural network forecasting model of slip zone loess seismic subsidence based on the data about slip zone loess dynamic tri-axial test,and make the error comparison with multivariate linear regression model.Results The assessments of BP neural network forecasting model are more accurate than multivariate linear regression model.Conclusion It′s a relatively optimum method used for forecasting,and with an important value to evaluation of the landslide stability in the loess land and analysis of the settlement of railway foundation.