| The technology of GPS has been used widely in structural health monitoring of large-scale civil engineering for its many advantages, such as all-weather work, fully automatic, fast efficient, high positioning accuracy and so on. As a small power weak signal, GPS signal in the propagation process is easily interfered by various external factors, such as multipath effect, cycle slips phenomenon and so on. These interference sources can cause a considerable proportion abnormal and distortion data, that leads to frequently error/disable alarm phenomenon in structural identification and fault diagnosis, therefore, whether in the information acquisition phase or in the safety evaluation phase, the data distortion and variation problems should be paid more attention in the GPS monitoring. According to the abnormally distortion test and identification problems in GPS data, the several aspects of theoretical and experimental research are as follows:(1) The relevant theory of GPS positioning technology and the classification of errors are systematically summarized. In order to decrease the abnormal or distortion data due to the measurement errors in GPS monitoring, the methods of reducing errors corresponding to the different errors arc presented.(2) Based on the GPS technology, the dynamic monitoring on the Dalian Beida bridge is completed. The finite element model of bridge is established by the ANSYS software. From the finite element model, the natural frequencies and modes of vibration arc provided to the layout of the rover stations. At last, the coordinates of rover station are solved by GrafNav/Net software, getting the real-time dynamic displacement of the bridge.(3) A new method for outlier detection of GPS monitoring data is proposed based on the control chart. Two kinds of control charts, Shewart control chart and cumulative sum control chart, are constructed and the early-warning models corresponding to two control charts are provided. Since the GPS monitoring data are not normally distributed, transferring them to Q statistic by kernel density estimation of cumulative distribution functions is raised, and based on this, the control chart of Q statistic used for GPS abnormal data detection is constructed. Finally, the detection capacity of Shewhart control chart and cumulative sum control chart are compared and analyzed based on the simulation and measured data. The results show that the Shewhart control chart is able to provide effective early-warning for the abnormal offsets3times the standard deviations, but it is lacking in the detection capacity of small offsets; while the cumulative sum control chart can accurately detect the continuous small offsets even smaller as0.5times the standard deviations, but the false alarm rate may be higher with the increase of the offsets. The different control chart can be chosen according to different practical engineering application.(4) The multi-step identification algorithm of GPS abnormal monitoring data based on the relational negative selection is proposed. Firstly, the relational model of GPS monitoring time series is established to determine the abnormal data initially and rapidly by setting the warning control limits. Then, the abnormal data is detected precisely for the second time by the proposed self-adaptive size radius negative selection algorithm. For the disadvantage of constant-sized detectors and self-representing methods, the corresponding improvements are proposed. Finally, the simulation and measurement data arc respectively used to verify the feasibility and efficiency of the method. The analysis results show that the negative selection algorithm with self-adaptive radius can cover larger non-self space by fewer detectors, which improves the abnormal detection rate of GPS data. Moreover, the proposed algorithm can accurately determine the extent of abnormal data, which has higher practical values in civil engineering. |
