地铁轨道结构沉降预测的神经网络模型

Research on the Settlement Prediction Models of Metro Track Structure based on Neural Network

地铁轨道结构的变形是影响地铁运营安全的重要因素之一,对其变形预测模型展开研究具有重要意义。本文以南京地铁2号线某区段的地铁轨道结构沉降监测实测数据为基础,研究分析了几种不同的沉降预测模型,并对预测效果进行了对比分析。论文首先介绍了时间序列模型之一,自回归模型AR(p);其次,介绍了神经网络BP模型,且确定地铁轨道结构沉降预测的BP模型结构为4×P×1。经工程实例分析,与时间序列模型相比,神经网络BP模型的预测精度能提高约50%,但该模型的缺点是模拟结果不稳定。最后,作者提出了时间序列与BP算法的融合模型,并详细介绍了该模型的具体结构和计算步骤。工程实例结果表明,融合模型的预测精度更高,与时间序列模型相比,精度能提高约60%,且融合模型的稳定性比常规BP模型要好。

The deformation of subway track structure is one of the important factors that affect the safety of subway operation,and it is of great significance to study the deformation prediction model.Based on the measured data of subsidence of metro tunnel structure of a section of Nanjing Metro Line 2,several different deformation prediction models are studied,and the prediction results are compared and analyzed.Firstly,the autoregressive model AR(p),one of the time series models,is introduced.Secondly,the BP model of neural network is introduced.The BP model structure for the settlement prediction of subway track structure is 4×p×1.Through the analysis of Engineering examples,the prediction accuracy of neural network BP model can be improved by about 50%compared with time series model.But the disadvantage of the model is that the simulation results are not stable.Finally,the author proposes a fusion model of time series and BP algorithm,and the structure of fusion model is introduced.The engineering example shows that the prediction accuracy of the fusion model is high.Compared with the time series model,the accuracy of the fusion model can be improved by about 60%,and the stability of the fusion model is better than that of the conventional BP model.