| Summary: | 碩士 === 立德管理學院 === 資源環境研究所 === 93 === This study proposed the detection model for outlier when processing timing data, pre-processed BAYTAP-G model and dynamic filter model to filter the known non-structural factors (such as tide, earth tide, atmospheric pressure, background interference and rainfall) in the groundwater timing data, and used outlier method to detect the structural factors (earthquake) that resulted in change of groundwater level.
The results of the outlier detection showed that earthquake and rainfall exist as IO outlier in the groundwater timing data, thus, having excellent detection effect on the time of occurrence of the unknown event. For earthquake without known time of occurrence, the two results provided inference and analysis different from physics viewpoints, and outlier method was used as reference to verify outlier or data for explanation.
This study also found that, in terms of the outlier detection model, it contains thoughtful and objective theories, but the dimension of the timing data and the characteristics of the data make it difficult for the analysis results to have reasonable explanation. Therefore, this study changed the inspection threshold accordingly to reduce the influence of the unknown factors in data, and attempted to reach a rational point between the theories and applications. Lastly, thresholds for different observation stations were proposed based on experience.
|
|---|
