Study On Application Of Space-time Series Analysis In Deformation Monitoring

Deformation analysis is a very important part of safety monitoring. It evaluates the safety of deformation bodies through the analysis of the deformation observation data and the establishment of a suitable mathematical model, which is of great significance to ensure the safety of the buildings, people’s lives and property.As the usual deformation analysis, it is mainly aimed at the single point to set up their own timing model, and doesn’t take into account the spatial correlation between the measured points. However, most deformation bodies work as a whole, their observation series both have time relation and spatial relation. Therefore, it is necessary to introduce the content of spatial analysis into the modeling of time series analysis, which is to model the sequence of Space-TimeSeries. This paper takes building settlement as example to make analysis and researching of space-time modeling. The main achievements are obtained as follows:1.The properties of the space-time series data are introduced in detail and the focus is on the analysis of spatial and temporal stability, temporal and spatial autocorrelation;2.Based on the analysis of time series, the modeling principle and steps of the spatial and temporal sequence model are mainly discussed by introducing the spatial adjacency and spatial weight matrix.3.Taking building settlement as an example, the model identification, parameter estimation and model checking are used to establish the STARMA model for the data series, and the applicability of space-time series modeling in deformation monitoring is studied and analyzed.4.Because most of the deformation data sequence is not smooth, and multiple difference may eliminate the useful information, as a result, the BP neural network and STARMA hybrid modeling of non-stationary sequence is proposed for such series. BP neural network is used to extract the temporal and spatial trends in the sequence, STARMA model is used for residual sequence, by the results of the example to show if the hybrid model has a better modeling ability for non-stationary data series.