基于神经网络与时空序列的混合模型在变形监测中的应用

摘要

In the area of deformation data analysis and modeling,space-time series model(STARMA)considers the time correlation as well as the spatial correlation among the points,which can describe the laws of deformation better,however,STARMA model is based on the linear stationary series,and most of the observed data series are non-stationary,so this brings limitation to the application of the space-time series model. As BP neural network has strong nonlinear mapping ability,in view of the above,combined with the characteristics of the two models,we construct SRATMA+ANN mixture model to deal with non-stationary sequences. According to the analysis of the settlement observation of buildings,the results show that the hybrid model is better than the single model and has a good practicality.%在变形数据分析与建模中,同时考虑变形体测点之间的时间相关性以及空间相关性的时空序列模型(STARMA)能够更好地反映出变形体的形变规律,但STARMA模型是建立在线性平稳模型基础上的,且大多数观测数据序列是非平稳过程,这给时空序列模型的应用带来了局限性.由于BP神经网络具有很强的非线性映射能力,基于此,结合这两种模型的特点,构造ANN+SRATMA混合模型来处理非平稳序列.通过对建筑物的沉降观测进行分析研究,结果表明了混合模型要优于单一模型,具有很好的实用性.