贵阳市地铁沿线InSAR地表沉降监测及GA-BP神经网络形变预测

InSAR Surface Subsidence Monitoring and GA-BP Neural Network Deformation Prediction along Guiyang Metro

摘要: 针对贵阳市城市建设中存在岩溶塌陷问题，基于SBAS-InSAR技术处理118景Sentinel-1A影像，获取贵阳市主城区2018年1月~2021年12月地表形变信息并与PS-InSAR监测结果进行交叉验证，基于此对研究区明显沉降区和地铁沿线150 m区域做沉降分析，最后用遗传算法改进的BP神经网络对沉降序列进行预测分析。结果表明：监测时间范围内，研究区地表沉降速率集中在−5 mm/a~1 mm/a，整体较为稳定，研究区无大范围明显沉降现象，存在部分明显沉降区，都存在山体开挖的现象，主要与人类活动有关。3条地铁沿线沉降整体稳定，每条线路存在1~2处明显沉降区；结合光学历史影像和城市规划资料分析，这些沉降主要与工程施工有关。改进的BP神经网络相较于标准BP神经网络有更好表现，绝对误差、均方根误差均为最小。

Abstract: Aiming at the problem of karst subsidence in the urban construction of Guiyang City, 118 scenes of Sentinel-1A images were processed based on SBAS-InSAR technology, and the surface deformation information of the main urban area of Guiyang City from January 2018 to December 2021 was ob-tained and crossed with the monitoring results of PS-InSAR. For verification, based on this, the settlement analysis of the obvious subsidence area in the research area and the 150 m area along the subway line is carried out, and finally the BP neural network improved by the genetic algorithm is used to predict and analyze the subsidence sequence. The results show that: within the monitoring time range, the surface subsidence rate in the study area is concentrated at −5 mm/a~1 mm/a, and the overall is relatively stable. There is no large-scale obvious subsidence phenomenon in the study area, and there are some obvious subsidence areas, all of which have the phenomenon of mountain excavation, mainly related to human activities. The settlement along the three subway lines is generally stable, and there are 1 or 2 obvious settlement areas in each line; combined with the analysis of optical historical images and urban planning data, these settlements are mainly re-lated to engineering construction. Compared with the standard BP neural network, the improved BP neural network has better performance, and the absolute error and root mean square error are the smallest.