【作者】 王芬；

【导师】 余明晖；

【作者基本信息】 华中科技大学 ， 系统工程， 2015， 硕士

【摘要】 在地铁施工过程中,为了实现安全管理和风险评估,获取准确的监测数据是非常有必要的。然而,由于地铁施工环境的复杂性、现有管理规范的约束以及数据采集方式为传统的人工采集等原因,导致直接采样的监测数据大多是不完整的或者有噪声的,因此不能直接用于风险评估与决策分析。分布式压缩感知理论(DCS),是在压缩感知理论的基础上发展起来的,实现对多信号的压缩与重构。与其他的重构方法相比,DCS的优势在于充分发掘了信号间和信号内的相关性来实现数据重构。本文主要研究在地铁施工过程中,基于DCS的地表沉降区域监测数据的联合重构。首先对地表沉降区域监测多信号的稀疏性和相关性进行分析,判断其是否满足DCS的适用条件。其次,利用DCS对地表沉降区域监测数据建立联合稀疏模型,并选择合适的重构算法。然后,利用所建立的联合稀疏模型对地表沉降监测数据进行离线重构与在线预测,并将实验结果与其他的重构方法进行对比分析。最后,利用DCS对随机采样时的地表沉降的采样下限进行估计。实验结果表明:对于地铁施工中的地表沉降区域监测数据缺失问题,基于DCS的数据重构方法要比回归分析法、稳健的主成分分析法(RPCA)、基于稀疏建模软件(SPAMS)方法的恢复精度要高。另外,当利用DCS对地表沉降监测数据进行数据重构时,联合的相关信号越多,重构误差越小。这正说明了DCS是通过利用信号间和信号内的相关性来提高信号重构精度的。随机采样时,利用DCS对地表沉降监测数据进行采样下限估计,根据给定的监测精度,预测结果要比实际采样频率要低。更多还原

【Abstract】 In the process of subway construction, it is necessary to obtain the accurate monitoring data for construction safety and risk assessment. However, due to the complexity of subway construction environment, the constraint of current management regulatory and the traditional way of manual data collection, most of the original data is incomplete or noisy, thus it cannot be directly used for risk assessment and decision analysis. Distributed Compressive Sensing(DCS), developed based on compressive sensing, is used for compression and reconstruction for multi-signal. Compared with other data reconstruction methods, the advantage of the DCS is that it makes full use of the correlations between inter-signal and intra-signal.This article mainly studies DCS-based field monitoring data joint reconstruction of ground settlements in metro construction. Firstly, the sparsity and relevance of the ground settlement field monitoring data is analyzed to determine whether it meets the applicable condition of DCS. Secondly, Joint Sparse Model(JSM) is built for the monitoring data by applying DCS, and an appropriate reconstruction algorithm is selected. Then the JSM is used for the offline reconstruction and online prediction of the ground settlement monitoring data. The experiment results are compared to regression analysis method, Robust Principal Component Analysis(RPCA) and Sparse Modeling Software(SPAMS). Finally, the sampling frequency of ground settlement is estimated based on DCS when sampling at random.The experimental results show that for the field monitoring data reconstruction of ground settlement in metro construction, the reconstruction precision of DCS is higher than regression analysis, RPCA, and SPAMS method. Additionally, the more relevant signals there are in the joint sparse model, the smaller the reconstruction error is. This convinces that the accuracy of signal reconstruction is improved by DCS due to the correlations between inter- and intra-signal is exploited. When sampling at random, according to a given monitoring accuracy, DCS-based sampling frequency estimation of ground settlements can provide a lower monitoring limit than actual sampling frequency.更多还原